Electronic copy available at: http://ssrn.com/abstract=2061393

The Adoption of Proportional Representation ∗

Lucas Leemann† and Isabela Mares‡

Abstract

The debate between economic and political explanations of the adoption of proportional rep-

resentation (PR) has occupied an important place in recent years. The existing tests of these

competing explanations have generated inconclusive results. We re-examine this debate and

argue that the causal mechanisms affecting politician’s decisions to reform the electoral system

operate at different levels of analysis. Reformulating Rokkan’s hypotheses, we show that both

partisan dissatisfaction with the translation of seats to votes and strong electoral competition

at the level of the district affect the decisions of politicians to support changes in electoral

institutions. In the empirical part, we evaluate the relative importance of (a) district level

electoral competition and vulnerability to the rise of social democratic candidates (b) partisan

calculations arising from disproportionalities in the allocation of votes to seats and (c) economic

conditions at the district level, more specifically variation in skill profiles and ‘co-specific in-

vestments’ in explaining legislators’ support for the adoption of proportional representation.

Article: p. 2-27References: p. 28-30Appendix: p. 31-47

∗ First draft: 3/20/2012; this version: 3/6/2013. We would like to thank Kate Baldwin, Pablo Beramendi,Matthew Carnes, Holger Döring, Miriam Golden, Shigeo Hirano, John Huber, Kimuli Kasara, Yotam Margalit, KevinMazur, Pierce O’Reilly, Susan Stokes, Pavithra Suryanarayan, Didac Queralt, Fabio Wasserfallen, Steven Wilkinsonand seminar participants at Princeton University, Lund University, and Sciences Po Paris for helpful comments andideas. Isabela Mares gratefully acknowledges financial support from the Center for the Study of Democratic Politics,Princeton University during writing this article.

† Department of Political Science, Columbia University; International Affairs Building; 420 W 118th Street; NewYork City; email: ltl2108@columbia.edu

‡ Department of Political Science, Columbia University; International Affairs Building; 420 W 118th Street; NewYork City; email: im2195@columbia.edu

1

Electronic copy available at: http://ssrn.com/abstract=2061393

1 The Adoption of Proportional Representation: One Phenomenon and a

Plethora of Explanations

The study of the process of reform of electoral institutions during the first decades of the twentieth

century and the adoption of proportional representation by some countries has been at the center

of comparative research in recent years (Boix, 1999; Benoit, 2004; Colomer, 2005; Blais, Dobrzyn-

ska and Indridason, 2005; Cusack, Iversen and Soskice, 2007; Calvo, 2009; Kreuzer, 2010; Cusack,

Iversen and Soskice, 2010; Renwick, 2010). This renewed interest in the origin of these political

choices has been partly motivated by a related literature examining the importance of proportional

representation in affecting a range of political outcomes, including the level of government expendi-

tures, the types of social spending programs, levels of inequality and so on (Lijphart, 1990; Huber

et al., 1993; Alesina et al. 2001; Person and Tabellini, 2004, 2005). While these studies have ad-

vanced a wide range of competing hypotheses about the economic and political determinants of

these electoral reforms, the empirical tests formulated to adjudicate among these competing expla-

nations remain inadequate. This empirical deficit arises because the core theoretical concepts of the

literature have been either measured at a level of aggregation that is, we believe, uninformative or

not measured at all.

Like any change in electoral rules, the adoption of proportional representation poses an immedi-

ate puzzle: why do politicians engage in risk-taking behavior and decide to replace the institutional

rules on the basis of which they have been elected? Existing explanations for the reform of elec-

toral institutions and the adoption of proportional representation cluster in two broad explanatory

families, stressing either economic transformations or changes in the broader political environment

as the source of political anxiety among right wing politicians. Let us consider the larger political

changes first.

Both hypotheses - stressing either the rise of Social Democracy or disproportionality in the al-

location of votes to seats - as determinants for demand for changes in electoral institutions can be

traced back to the work of Braunias (Braunias 1932, see also Rokkan, 1970: 157-58). Contempo-

rary studies seeking to explain the adoption of proportional representation have tested one of the

two Rokkanian hypotheses. Boix (1999) has provided a cross-national test of the first Rokkanian

hypothesis. In recent years, Calvo (2009) has revived this second Rokkanian explanation and has

provided simulations informed by comparative cases to illustrate this logic.

2

In Rokkan’s original study, the two political hypotheses acted as substitutes. Rokkan conjectured

that in some countries - such as Germany, Norway or Sweden - the socialist mobilization played a

significant role in accounting for the adoption of proportional representation, while in other coun-

tries - such as Belgium, Denmark or Switzerland - considerations about the disproportionality of the

translation of seats to votes were the primary factor in accounting for the motivation of politicians

to adopt changes in electoral institutions (Rokkan, 1970: 157). In this paper we argue, however,

that the two political hypotheses are complements, rather than substitutes. Both explanations can

operate within the same case, but capture constraints on the activity of legislators that operate

at different levels. The first political explanation locates the salient constraint at the level of the

electoral district, in district-specific electoral vulnerabilities encountered by candidates on the po-

litical right that face social democratic challengers. By contrast, the second Rokkanian explanation

locates the salient constraint at the level of the party. An important task for the analysis is to

identify the relative importance of district specific vulnerabilities and partisan variables in affecting

the preferences of legislators for changes in the design of electoral institutions.

The alternative, economic perspective has been articulated by Cusack, Iversen and Soskice

(2007).1 According to this approach, “the choice of electoral systems is endogenous to the struc-

ture of economic interests” (Cusack et al. 2007: 388). The key structural economic conditions that

are hypothesized to generate differences in political preferences and variation in political outcomes

(adoption versus non-adoption of proportional representation) are the presence of ‘co-specific as-

sets’, in other words, investments in human capital that are jointly made by employers and workers

(Cusack et al. 2007). Cusack et al. hypothesize that in economies characterized by ‘consensual la-

bor relations’ and high levels of ‘co-specific’ assets, economic actors consider majoritarian electoral

rules as potentially detrimental to their long-term economic interests. First-past-the-post electoral

systems allow groups with a narrow geographic base to capture political power and enact legislation

that could potentially harm actors that had made long-term economic investments. In the presence

of significant investments in skills, Cusack et al. predict that the most significant interest groups

and the political parties to which they are tied will demand the adoption of proportional represen-

tation. By contrast, majoritarian institutions persist in economies with lower levels of co-specific

assets. In contrast to the first family of explanations, economic hypotheses about the adoption of1See Rogowski (1987) for an economically motivated alternative explanation of PR adoption. In this explanation

it is not the skill level but the degree to which a country is involved in international trade.

3

proportional representation derive their political predictions about the demand for changes in elec-

toral institutions from the distribution of economic endowments and less from partisan or electoral

considerations.

All explanations, political or economic in nature, are framed at a theoretical level that is quite

abstract. Ultimately, they seek to explain choices made by individual legislators. As a consequence,

many of these persistent disagreements cannot be resolved with the existing cross-national data.

Additional leverage can be gained, however, by shifting to a lower level of empirical aggregation,

the individual politician. This move to the individual level is desirable for a number of reasons.

First, the dominant empirical strategy pursued in contemporary research which is premised

on quantitative analysis of variation among a small number of countries is severely constrained

by the small number of observations. The number of observations in cross-national studies is

relatively low, varying from 24 in the original Boix paper of 1999 to 12 in some of the specifications

presented in Cusack et al. (2007). This small number of cases, coupled with a high multicollinearity

among competing variables has limited our ability to assess the relative importance of competing

explanations. This constraint might also explain why so far scholars have been unable to test all

three explanations of the adoption of proportional representation, but have tested only one or two

theories at a time. Second, an analysis that is situated at the individual level is also preferable

because some of the central predictions of the first political hypothesis – stressing the importance

of electoral vulnerabilities – are located at the level of the electoral district. Existing studies that

have relied on cross-national data have measured many of the relevant variables capturing political

competition at a level of analysis that is too aggregated and, thus, uninformative (Cox, 1997).

The difficulty and real challenge for an empirical evaluation of competing explanations of PR

adoption using roll call data is the development of measures of the core explanatory variables of

the economic explanation, the level of ‘co-specific assets’. We take up this challenge in this paper

and develop precise estimates of the ratio of skilled workers at the level of the electoral district.

This allows us bring competing explanations about the adoption of proportional representation to

a common empirical unit of analysis and assess, for the first time, the relative effect of structural

economic conditions at the level of the district, electoral competition and partisanship on individual

level support for the adoption of proportional representation.

4

Our paper examines a decisive vote taken by the German Reichstag as a result of which Ger-

many adopted proportional representation.2 The German case occupies a central position in all

explanations that have been advanced to explain the adoption of proportional representation. Con-

sider first its importance for political explanations of changes in electoral institutions. The existing

literature treats Germany as the paradigmatic case that illustrates the importance of the Social

Democratic threat as the catalyzing factor leading to a change in the position of right wing parties

over the desirability of proportional representation. This interpretation of the German case goes

back to the classic studies of the adoption of proportional representation, such as Braunias (1932)

or Rokkan (1970). As Rokkan argued, in countries where “hostility and distrust” among existing

parties on the right was too strong, the latter found it impossible to make common cause against

the socialist threat. As such, they preferred proportional representation, a system that guaranteed

them the ability to compete independently (Rokkan 1970: 158). This account, however, has dif-

ficulties in explaining the diverging positions taken by different parties on the political right over

the question of electoral reform change. While the catholic Zentrum and the German Conservative

Party opposed the adoption of proportional representation, both the National Liberals and the Free

Liberals supported the adoption of changes in electoral institutions. Thus, it appears that both

political explanations seem to capture important determinants of the motivations of politicians to

support changes in electoral institutions, but so far no study has attempted to disentangle the

relative explanatory power of these two explanations.

Germany is also the case that motivates the economic explanation of the adoption of pro-

portional representation. The extensive network of vocational training (Thelen, 2004) that was

well-established during the 19th century led to the development of a workforce characterized by

high ratios of skilled workers and an unusually high level of vocational skills (Kaiserliches Statistis-

ches Amt 1899: 73; Bade, 1980). The central hypothesis advanced by Cusack, Iversen and Soskice

suggests that the regional variation in the distribution of human capital has important political

consequences both for the adoption of proportional representation and for democratization more

broadly. Germany occupies, thus, an asymmetric position for economic and political explanations

and represents a ‘hard’ case for Rokkanian explanations (George and Bennett, 2005). Due to this

asymmetry, empirical results that disconfirm importance of human capital as a predictor of the2This represents the only recorded vote on the issue of proportional representation taken by the German Reichstag

during the period between 1870 and 1918.

5

adoption of proportional representation in the German case can weaken our confidence in the com-

parative explanatory power of the economic approach to the origin of proportional representation.

The following section explores the tensions among the two political hypotheses concerning the

determinants of support of changes in electoral institutions and formulates both the hypotheses and

the empirical strategy used to assess the relative importance of partisan considerations and district

level electoral competition on legislator support for changes in electoral institutions. We turn next

to the economic explanation, discuss its empirical predictions and then discuss our measures of

human capital distributions across electoral districts.

2 Economic Conditions, Political Competition and the Change of Electoral

Institutions

Stein Rokkan’s shadow looms over much of the contemporary research seeking to explain the adop-

tion of proportional representation in Europe during the first decades of the twentieth century

(Rokkan, 1970). Two of the dominant hypotheses concerning the adoption of proportional represen-

tation can be traced back to his work. The first of these argues that incumbent elites support the

adoption of proportional representation to avoid a complete electoral disaster in the face of social

democratic mobilization (Rokkan, 1970). Boix takes up this hypothesis and argues that if parties

on the political right are equally balanced in their electoral strength, but unable to coordinate on

a common candidate (either at the level of the constituency or nationally), then they are likely

to support the adoption of proportional representation (see Blais et al., 2005). By contrast, the

incentives to adopt changes in electoral institutions are weaker if one party dominates the elec-

toral arena (Boix, 1999, p.612).

Rokkan’s second hypothesis – taken up in current research by Ernesto Calvo (2009) – attempts

to specify conditions under which incumbent parties on the right favored changes in electoral insti-

tutions even in the absence of social democratic threats. This hypothesis suggests that parties with

a geographical dispersed distribution of votes also faced incentives to demand changes in electoral

institutions, as multiparty competition increased the seat-vote disproportionality of the electoral

results. According to this logic, the adoption of proportional representation is regarded as a mech-

anism to reduce the “severity of the partisan bias and of electoral regimes to districting problems”

(Calvo, 2009, p.256).

6

The two Rokkanian hypotheses differ in the relative importance placed on the social democratic

threat. Both hypotheses stress, however, ‘electoral coordination’ (or failure thereof) of incumbent

parties as a critical variable influencing the decision of incumbents to support or oppose changes in

electoral rules and the adoption of proportional representation. Critics have noted that the notion

of electoral coordination is theoretically unclear and poorly operationalized empirically. Cusack et

al. note two limitations of the test of electoral coordination presented in Boix (Cusack, Iversen and

Soskice, 2007, p.375). The first is the failure to take account of the opportunities for coordination

created in electoral systems that used runoffs. Runoffs allowed parties on the political right that

competed in the first round to form electoral alliances in the second round to oppose a social

democratic candidate. Cusack et al. note some limitations of the empirical test of the Rokkanian

notion of electoral coordination employed by Boix and argue that the variable used by Boix, the

“effective number of parties” fails to distinguish between cases in which one party on the right is

dominant from cases where parties of the right are equally sized (Cusack, Iversen and Soskice, 2007,

p.375). A similar critique is presented by Blais, Dobrzynska and Indridason (2005) who argue that

the incentives of all parties to switch to proportional representation are stronger in countries which

require a majority rather then just plurality of the votes to take a district.

While raising these criticisms, Cusack et al. do not propose a superior empirical solution. It is

impossible to measure district level electoral competition with cross-national data. Using aggregate

data we cannot learn whether or not there was coordination on the right. Clearly, runoffs could

alleviate potential coordination problems among parties on the right. However, while runoffs induce

some electoral coordination during the second round, they do not necessarily produce electoral

coordination among right wing parties (Bertram, 1964; Fairnbairn, 1990; Reibel, 2007). Parties on

the political right have a range of possible strategies during runoffs. These include the formation

of pre-electoral coalitions which are restricted to right wing parties, political support for the left

wing candidate over the right wing opponent (in districts where a candidate on the right encounters

a candidate on the left during runoffs) or recommendation to their voters to abstain during the

second round (and their de facto refusal to endorse any of the remaining candidates). We provide a

number of examples in the appendix to illustrate how abstention recommendations contributed to

the Social Democratic victories (see section A1).

Studies emphasizing the economic foundations of proportional representation do not derive the

demand for the adoption of proportional representation from partisan expectations about their

7

electoral survival, but from the expectations of leading economic groups (employers and labor) about

the different regulations that might emerge under different electoral rules. Thus, what distinguishes

the trajectory of PR from non-PR adopters is the level and depth of investment in skills made

during the early stages of industrialization. As Cusack et al. formulate their hypothesis linking

specific assets and proportional representation:“Two broad alternative patterns of labor and skill regulation in industry can be found at thiscritical period in the early twentieth century. They relate to the question of the control ofskilled formation and the content of skilled jobs. The first possible alternative, which we findsubsequently in each of the PR adopters, was some form of cooperative agreement between busi-ness and unions: in exchange for collective bargaining rights and monitoring of skill formation,business would have managerial control of the shop floor and determine training levels. Thisalternative had major implications for the politics of regulation because skills in this system areco-specific assets. [...] Unions and employers become vulnerable to opportunism and holdup andthey consequently need credible long-run guarantees which include an appropriate frameworkagreement at the political level to underwrite the relevant labor market and social security in-stitutions and rules. The political system has to be such that the agreement cannot be changedby a change of government without the consent of other groups. This requires not just a systemof proportional representations to enable the different groups to be represented through parties,but also a political system that allows for consensus decision making in the regulatory areasthat concern them” (2007: 328)

We interpret the argument as generating two testable empirical implications. The first is that

in countries or regions where investments in specific skills are higher, economic actors seek to

establish proportional systems of representation either independently or through their partisan

representatives. By contrast, the demand for proportional levels of representation is lower in regions

with lower levels of co-specific assets. Secondly, regions characterized by high levels of investment

in skills also experience more consensual relations in the industrial sphere and among parties.

Yet, despite the straightforward nature of this empirical implication, Cusack et al. have not yet

provided a direct empirical test of their theory. In their cross-national analysis, the variable used

to test the importance of ‘co-specific assets’ is a composite variable that includes the strength of

employer associations and the existence of rural cooperatives and pre-existing guilds. The other

indicator, the strength of employer associations is a problematic empirical measure. As Kuo (2010)

has illustrated, most employer associations that existed in Europe prior to World War I (and thus

prior to the adoption of proportional representation) were established with the explicit goal to

repress labor movements. The existence of a coordinated employer association does not proxy for

harmonious labor relations and co-specific investments, the critical variable of this theory.

The most recent APSR paper which restates the asset-specific explanation and offers limited

qualitative evidence discussing electoral reforms in Imperial Germany does not provide any direct

8

measures of human capital development across German regions. Saxony, Württemberg and Baden

are used interchangeably as being, presumably, cases that share similarly high endowments of human

capital (which is the hypothesized factor leading to the adoption of proportional representation).

Cusack et al. provide no empirical evidence to support these conjectures. At the turn of the century,

Baden and Württemberg are relatively underdeveloped regions of the German Empire, displaying

levels of agricultural employment that are higher than the respective levels in the Reich (Hohls and

Kaelble, 1989, p.167- 187).3 Our estimates of human capital development – which computes ratios

of skilled workers – present roughly similar patterns. The national average is 36.2% (sd: 9.0) and

while Saxony reaches 52.7% (sd: 3.01) we find that Baden has 36.8% (sd: 2.8) and Württemberg

is slightly below average with 35.8% (sd: 1.47). These measures do not support the argument that

all three regions (Saxony, Baden, and Württemberg) share high endowments in human capital (see

also Fremdling and Tilly, 1979). Thus, qualitative assessment of 19th century economic realities

(which are likely to be based on 20th century outcomes) turn out to be inaccurate. Given that the

core explanatory variable (distribution of ‘asset specific investments’) takes different values across

these regions, it cannot serve as the common explanation for subnational electoral reforms.

While the debate between economic and political explanations of the adoption of proportional

representation has occupied a visible place in prominent publications, it has not yet been resolved.

Quite the contrary. This inconclusiveness has several reasons. Existing explanations have been

brought to the data only imperfectly. Ambiguity exists as to what constitutes electoral coordination

and how it should be operationalized. Many of the existing tests of the first Rokkanian hypotheses

have only assessed the effects of variation in national level competitiveness on the probability of

transition to proportional representation. This remains both an imperfect test of the first Rokkanian

hypothesis which has to be tested using more disaggregated subnational and district level measures.

The empirical assessment of economic explanations is a necessary step in the evaluation of competing

explanations for the adoption of proportional representation. For this endeavour, the lack of a

measure of the core explanatory variable, the co-specific assets, is an outstanding difficulty.3At the time of the 1895 census, the share of the labor force employed in agriculture in Baden and Württemberg

stood at 44 and 45%, respectively. The average agricultural share for the entire Reich stood at 37%. By contrast,Saxony’s share of agricultural employment stood at 17%. These relative levels of economic development did notchange much until 1905-1907, the period when the last economic census prior to World War I was assessed. Theemployment shares in agriculture in Baden declined from 45 to 40% but remained relatively stable in Württembergat 45%. In both regions, levels of industrial development continued to be below that of the Empire (which stood at35% in 1905). By contrast, Saxony’s share of agricultural employment stood at 13% in 1905. The figures are basedon the aggregate results of both economic censuses presented in Hohls and Kaelble (1989: 167- 208).

9

3 A Theoretical Re-Formulation of the Rokkanian hypotheses

The two political hypotheses locate the prime determinant of legislator’s calculation about the

attractiveness of different electoral rules at different levels of analysis, the individual district versus

the party. The first hypothesis highlights electoral vulnerabilities at the level of the district. In this

hypothesis the general contestation of a race, the availability or absence of coalition partners, the

strength of the contender on the left and the potentially scarring experience of a runoff are likely

to affect the decision of a politician to support changes in electoral institutions. By contrast, the

second political hypothesis locates the important determinant at the level of the party and not the

district. Partisan preferences over the desirability of different electoral rules are influenced by the

attractiveness (or its absence) of the way in which the policy status quo allocates votes to seats.

In their calculation about the relative attractiveness of different electoral rules, politicians are

constrained both by district-specific factors and by the general considerations of their party. The

partisan and district level incentives can reinforce each other for support of particular electoral rules

but they can also come into tension. Consider the case of a politician representing a party from the

right that experiences an unfavorable distribution of votes to seats and, who has been elected by

a narrow margin in a runoff against a social democratic opponent. For this particular politician,

the constituency level incentives and the partisan incentives reinforce each other and one expects

this politician to support the adoption of proportional representation. Partisan and district-specific

electoral incentives may however come into tension.

By hypothesizing that the two political explanations about the origin of proportional represen-

tation highlight political incentives that are located at different levels of analysis, we challenge a

number of statements made by Rokkan. Instead of viewing the two political hypotheses as separate

roads to proportional representation undertaken by different countries, we argue that both logics

can be at work within the same country. The two hypotheses specify different constraints on the

behaviour of legislators. To formulate this point more starkly, there are no two roads to proportional

representations, but two distinct sets of political constraints on the behaviour of politicians, which

operate at the level of the district and party respectively. A legislator-based account of the origin

of proportional representation allows us to disentangle the relative effects of these constraints.

10

3.1 District-Level Vulnerabilities versus Partisan Constraints

To test the implications of the first Rokkanian hypothesis, we examine the consequences of district

specific factors on the probability of support of changes in electoral institutions. We hypothesize

that three variables linked to the contestability of a race are likely to affect the support of the

legislator for changes in electoral institutions. The first is whether the politician is elected in the

first round or a runoff. All things equal, we hypothesize that politicians elected in runoffs are more

vulnerable than politicians elected during the first round. Runoffs are themselves an indicator of

coordination difficulties for parties on the right. As such, we hypothesize that politicians on the right

elected during runoffs are more likely to support changes in electoral institutions than politicians

elected during the first round. A second district-specific factor that is likely to affect preferences

towards electoral institutions is the margin of victory. All things equal, politicians elected in races

with larger margins are more likely to support the policy status quo than politicians elected in

more contested races with narrower margins. A final prediction of the first Rokkanian hypothesis

is that the magnitude of the threat of the social democratic party (as measured by the vote share

of the Social Democratic candidate during the first round) is likely to increase support for changes

in electoral institutions.

We include three variables to test for these hypotheses linking electoral vulnerability in the

district and preferences for changes in electoral institutions. They include a variable that takes the

value 1 when the politician was elected during runoffs, a measure of the margin of victory during

the decisive electoral round and a measure of the vote share of the Social Democratic candidate.

In combination, these variables allow us to capture only the theoretical expectations laid out by

the first Rokkanian hypothesis. They capture district-specific electoral constraints. They remain

insufficient in capturing the full set of political constraints and considerations facing the politician.

To understand the latter, we need to consider partisan considerations about the effects of the

electoral formula on the electoral fate of the party. We turn to a discussion of these considerations

next.

3.2 Partisan Preferences about Changes of Electoral Rules

The second political hypothesis formulates its predictions at the level of the party rather than

district. In this hypothesis, the key variable that is likely to affect the preferences of parties about

11

the adoption of proportional representation is the disproportionality in the allocation of seat to

votes under the existing electoral rules. Parties that benefit from the current allocation of seats

to votes (by obtaining a higher seat than vote share) are likely to favor the policy status quo. By

contrast, parties that lose from the allocation of seats to votes are likely to favor changes in electoral

institutions and the adoption of proportional representation.

Table 1 presents descriptive information on the effects of disproportionality for the major political

parties during the 1912 election of the Reichstag, which elected the deputies that ultimately adopted

the legislation introducing proportional representation. We group existing deputies in six partisan

families, which include Social Democrats (who at the time obtained the largest number of votes),

the Zentrum, National-Liberals, Free Liberals, Conservatives, and parties representing minorities

(such as Poles or the Elsass independents). Column 2 presents information on the number of votes

cast for these candidates, column 3 on the number of seats held by the respective party during the

last session of the Reichstag. Column 4 presents the measure of seat-vote disproportionality for the

party and column 5, the prediction about support for changes in electoral institutions.

Table 1: Seat-vote disproportionality for the largest parties in the Reichstag during the 1912 election

Party Vote Share Seat Share Disproportionality Prediction of Rokkan II

SPD 34.8 % 27.7 % -7.1 Support of PR

Zentrum 16.8 % 22.9 % 6.1 Opposition to PR

National Liberals 13.6 % 11.3 % -2.3 Support of PR

Conservatives 11.6 % 14.1 % 2.5 Opposition to PR

Free Liberals 13.1 % 10.6 % -2.5 Support of PR

Minorities 4.5 % 8.3 % 3.8 Opposition to PR

Source: Reibel (2007) based on Kaiserliches Statistisches Amt (1913: 4-103) and www.wahlen-in-deutschland.de.

As Table 1 illustrates, the Social Democratic Party experienced the most disfavorable ratio in

the allocation of votes to seats. Industrialization which led to the concentration of voters in urban

areas aggravated the disproportionality experienced by the Social Democratic Party (Boix, 2010).

The disproportionality worsened from (−2.6) in 1871 to (−7.1) percent in 1912. As a result, the

Social Democratic Party became the staunchest advocates for the introduction of proportional rep-

resentation. During the period leading to World War I, Social Democratic candidates cosponsored

legislation recommending the introduction of proportional representation in national elections for

12

the Reichstag and actively advanced similar proposals during discussions of electoral reforms of the

subnational parliaments Schmädeke (1995). Several studies assume that Social Democratic parties

were supportive of the adoption of proportional representation. As recent studies have pointed

out, this assumption may be inaccurate and cannot adequately account for the observed empirical

variation in the preferences of Social Democratic parties across Europe on the question of electoral

reform (Ahmed, 2010; Penadés, 2008). We can, however, derive theoretically the preferences of

Social Democrats by taking seat-vote disproportionality into account.

Among the parties on the right, the two prominent losers in the allocation of votes to seats were

the Free Liberals and National Liberals. The disproportionality index during the 1912 elections

stood at (−2.5) for Free Liberals and at (−2.3) for National Liberals. In contrast to Social Democrats

that had been disfavored by the allocation of votes to seats during all elections of the period, for the

two parties on the right the disproportionality in the allocation of votes to seats was a more recent

phenomenon that came into place during the last elections of the Reichstag (Duverger, 1963: 375).

Party leaders from these two parties embraced the idea of proportional representation later than

the Social Democrats and the question about the advantage of this electoral reform remained the

subject of significant intra-party disagreement (Gagel, 1958, p.125-141). For Free Liberals, the push

for the adoption of proportional representation originated with party leaders from Southern German

states where the presence of catholic minorities in some districts could lead to the waste of votes won

by the Free Liberal candidates under the existing single member electoral system (Simon, 1969).

The two large political parties on the right that benefited from the allocation of seats to votes

were the Zentrum and the Conservatives. Party leaders from both parties understood the electoral

advantage of existing electoral institutions and rejected the introduction of proposals calling for

the adoption of proportional representation (on Zentrum: Penadés, 2011; on Conservatives: Kühne,

1994). This opposition to any reform of the electoral system remained one of the issues that brought

these two parties – which had been bitter enemies during the Kulturkampf – together and led to the

creation of the ‘black-blue’ electoral alliance during the 1912 election (Bertram, 1964; Reibel, 2007).

An internal report commissioned by the German Interior Ministry (Reichsamt des Innern) in

the immediate aftermath of the 1912 election noted the divergent rates of success of different parties

of the right in translating their votes into seats. The report observed that while the “Liberal Party

obtained over 3 million votes in 1912, it was able to obtain only 4 seats in the first round, while

13

the Zentrum obtained over 80 seats with only 2 million votes. The explanation for this fact is”,

the report noted, “that there are no areas where there are only Liberals; rather there are areas

where there are Liberals and either Conservatives, or Zentrum or Social Democrats” (Bundesarchiv

Berlin Lichterfeld R1501/192-193). Prominent liberal politicians also noted the worsening seat vote

disproportionality encountered by their party and began to advocate the adoption of proportional

representation (Brandenburg 1917: 30, Gagel 1958: 128).

The Social Democrats had militated for changes in electoral institutions beginning with the 1891

Erfurt congress. By contrast, Free Liberal and National Liberal politicians began to advocate the

adoption of proportional representation only much later (Gagel, 1958). This shift in the position

on the desirability of electoral reform advocated by these two parties on the right guaranteed the

formation of an encompassing political coalition supporting the adoption of proportional represen-

tation. The political factor that motivated this change in strategy among these two parties on the

right was the worsening of their seat-vote disproportionality.

Let us consider how considerations about the translation of votes to seats affected the policy

positions of a range of parties on the question of electoral reform. From the perspective of the Social

Democratic Party, calculations of their expected vote share under this alternative electoral rule (that

used as input the number of votes received during the 1907 election) revealed an expected gain of

76 seats (from 43 to 117 seats) (Andersen, 2000: 343; Fricke, 1961). Both National Liberal and

Free Liberal Parties that experienced losses in the translation of votes to seats under the existing

electoral rules also favored the adoption of proportional representation (Andersen 2000: 343, Gagel

1958: 143). By contrast, the two other parties on the right that gained from the allocation of the

votes to seats under single-member electoral rules – the conservatives and the Zentrum – opposed

changes in electoral rules. A recommendation of the Ministry of Interior summarizes the fear of

Conservatives about changes in electoral rules. The adoption of proportional representation, this

study showed, would reduce the seat share of conservative parties by 8 seats (as compared to the

1912 results) and “cannot be regarded as a cure against the existing inequalities in representation”.

The expected loss of seats for the Zentrum, as compared to its 1912 electoral results, stood at 32

seats (Bundesarchiv Berlin Lichterfeld R1501/343).

We have begun to explore the implication of the second Rokkanian logic for the preferences on

reform of electoral institutions. Parties losing from the allocation of votes to seats (Social Democrats,

14

National Liberals and Free Liberals) supported the introduction of proportional representation. By

contrast, winners of vote seat disproportionality (Conservatives and Zentrum) opposed it. How

constraining were these desiderata of party leaders on the decisions made by individual legislators

during the actual vote of electoral reforms? Our empirical analysis allows us to explore these

questions and examine whether individual legislators internalized these needs of their respective

parties and of the relative effect of district specific vulnerabilities in affecting their calculations. To

test how the disproportionality in the allocation of votes to seats affects the preferences of party

leaders, we include the measure of disproportionality (SEAT VOTE DISPR.) as additional control

in our analysis of the vote choice of legislators.

3.3 Asset Specific Investments and the adoption of proportional representation

Both Rokkanian explanations are political explanations. They locate the source of demand for

changes in electoral institutions in either electoral vulnerabilities at the district or in partisan

calculations. The causal chain presupposed by these hypotheses is relatively short: politicians, it is

hypothesized, respond either to their increased electoral vulnerability or to calculations about the

translation of votes into seats. By contrast, economic explanations locate the source of political

demand for electoral change in economic conditions at the district level. The causal chain linking

these economic factors the actions of legislators is long and tenuous. In this alternative explanation,

politicians favor the adoption of proportional representation not because it serves an immediate

political need but because they expect that proportional representation will produce more favorable

labor regulation that will protect the ‘asset specific investments’ made by employers and firms in

their districts (Cusack, Iversen and Soskice 2007 and 2010). Let us suspend possible questions about

the farsightedness of politicians and assume that this logic captures plausible political constraint on

the activity of legislators (e.g. Andrews and Jackman, 2005). The empirical question that we seek

to answer is whether structural conditions at the district level affect the political choices made by

legislators once we control for all the other relevant political factors (Rokkan 1 and Rokkan 2) that

presuppose a much shorter causal chain.

In section 2, we have noted that non-Rokkanian approaches have not yet tested the core ele-

ment of their explanation – which hypothesizes that support for proportional representation can be

predicted by high levels of ‘asset specific’ investment. We seek to remedy these shortcomings and

15

replace the imprecise characterizations of variation in ‘co-specific assets’ with actual measures of

human capital development across German electoral districts. While difficult and computationally

intensive, it is possible to compute measures of the distribution of human capital across German

regions prior to the adoption of proportional representation.

We use two distinct sources of the German occupational census to compute this measure. First,

we start with tables of the ratio between skilled and unskilled workers. As part of its 1895 census,

the German Statistical Agency has collected information on the ratio between skilled and unskilled

workers in 183 occupations, which includes 161 industrial occupations and 22 occupations in ser-

vices. The definition of skilled workers (“gelernte Arbeiter”) employed by the Statistical Agency

is straightforward (but somewhat laconic): “skilled workers (gelernte Arbeiter) are those workers

that require some training. By contrast, unskilled workers (ungelernte Arbeiter) – in other words

manual workers, handymen and other workers in services – which do not have existing training”

(Kaiserliches Statistisches Amt 1899: 73). The study noted that the number of skilled workers is

very high in Germany (nearly 50% of workers employed in industry). “The economic relations in

industry – which requires particular abilities and skills which can be learned or acquired through

persistent exercise are the decisive factor which explain these high ratios of skills” (Kaiserliches

Statistisches Amt 1893: 73). We present the disaggregated levels of skills for all 161 occupations in

the appendix (Table A8, p. 42) and provide further information on this measure in the appendix

(see section A2).

Our second input is Germany’s occupational census. In addition to the information on skill

profiles of occupation, the 1895 occupational census (Gewerbezählung) contains extremely disaggre-

gated information on the employment shares across these 180 occupations in 1003 German localities.

These two sources allow us to estimate the ratio of skilled to unskilled industrial workers across each

locality. In a final stage, we aggregate localities at the district level to obtain measures of the density

of human capital across the 397 electoral districts of Imperial Germany. Using the number of skilled

workers in each district as an input, we compute two measures of skill profiles that differ in their

respective denominator. The first is a measure of skills as a percentage workers employed in industry

(SKILL). The second is a ratio of skilled workers to all employed persons in the district (SKILL

LEVEL). Finally, we also use a measure of skills aggregated at a higher level which corresponds to

the 29 subnational political units (SKILL REGION). We present the results using the second and

third measure of skills in the appendix of the paper (Table 5, Table 6, Table 7, and Table 8).

16

4 Empirical Analysis: Explaining Individual-Level Support for the Adoption of

Proportional Representation

The dependent variable in our analysis is the support for a bill recommending the adoption of pro-

portional representation and “the creation of larger electoral districts” (Stenographische Berichte

des Deutschen Reichstages, 13. Legislaturperiode, Volume 323, p. 5962-5964). The bill was ap-

proved by a rather narrow political margin of 10% points, with 55% of deputies voting in favor,

45% deputies opposing it and 3 casting invalid ballots. The vote in the Reichstag represents the

endpoint of an intense (but relatively short) period of parliamentary negotiations of changes in

electoral institutions for elections to the second chamber of the national parliament. We conducted

an extensive search of all legislative bills recommending the introduction of proportional representa-

tion throughout the 13 legislative sessions of the Reichstag between 1870 and 1912, but only found

legislative initiatives recommending the adoption of proportional representation during the 13th

Reichstag that was elected in 1912.

4.1 Empirical Evidence for Motivations to Change Electoral Systems

We attempt to solve the puzzle presented by the existing literature on the origin of proportional

representation; competing explanations of this political choice differ widely in their postulated causal

mechanisms. Our test is decisive in adjudicating among these competing explanations for a number

of reasons.

First, it has the crucial advantage that it is carried out at the appropriate level of analysis,

the individual level. Since the decision to change an electoral system is made by a politician

representing a district we should be relying on district-specific measures; we can use the margin

of victory in a district instead of relying on a country-wide average to gauge the closeness of vote

outcomes. Secondly, it allows us to test for a range of additional factors –most notably the partisan

concerns about disproportionality – that have not yet been incorporated into a economic models

explaining electoral system change. Finally, our measures of the theoretical concepts invoked by

various explanations (such as asset specific investments) are, we believe, more precise than the

existing indicators of coordination used in cross-national research. The latter measures are both

quite distant from the core variables that are hypothesized to affect the outcomes (‘skills’) and also

susceptible to measurement error (Kreuzer, 2010).

17

We carry out a full test that includes the salient variables in each of the main hypotheses in

the literature, including the second Rokkanian hypothesis. We use predictors originating from all

three political explanations: district specific vulnerabilities and socialist threat (Rokkan I), vote-

seat disproportionalities (Rokkan II) and levels of skills acquired through the vocational training

system (Cusack, Iversen and Soskice, 2010). This is not only the first test that includes all the

variables but also the first time that these hypotheses are tested at the district level, which is the

correct level for an empirical test.

The models presented in Table 2 and Table 3 operationalize all three hypotheses. Our test of the

second Rokkanian hypothesis is the degree of disproportionality in the translation of votes to seats.

We use the dataset on electoral politics in Imperial Germany (ICPSR 1984) which we cross-check

with Reibel (2007) to determine the partisanship of the winning politician in each district.

The measure of disproportionality (SEAT VOTE DISPR) takes higher values if the share of seats

held by each party is higher than its share of votes. Thus, a positive number reflects a party that

gains from the existing electoral arrangement. Negative numbers indicate parties whose fortunes

might improve under proportional representation (see Table 1). We use three variables to test for

the effects of the district specific uncertainties postulated by the first Rokkanian hypothesis on the

decision of legislators to support proportional representation: a measure of the strength of the Social

Democratic party in a district (SOCIALVOTE), a measure of the decisive electoral round which

takes the value 1 if the politician is elected during the second round (SECOND), and a measure of

the closeness of the race among non-socialist parties in the first round (RIGHT MARGIN).

Three additional explanatory variables model differences in economic and social conditions across

districts. We use our skill measure (SKILL) presented in Section 3.3 to account for the consequences

of different ’investments in co-specific assets’, the critical explanatory variable proposed by Cusack,

Iversen and Soskice. Our measure of differences in economic development is a variable that captures

the share of the labor force that is not employed in agriculture (NON AGRICULTURE). We also

measure the share of catholics (CATHOLICS) in a district to account for the possible effects of

religious heterogeneity. This measure is consistent with cross-national analyses of the effects of

religious fragmentation with the probability of adoption of proportional representation (Boix, 1999).

18

Table 2: Probit Models on PR VoteModel 1 Model 2 Model 3 Model 4

Partisan SEAT VOTE DISPR. -0.301*** -0.391*** -0.406*** -0.414***(0.037) (0.056) (0.058) (0.061)

District SOCIALVOTE 0.026** 0.026** 0.111** 0.100**(0.011) (0.013) (0.044) (0.047)

District SECOND*SOC.VOTE -0.095** -0.093**(0.039) (0.043)

District RIGHT MARGIN*SOC.VOTE -0.001 -0.001*(0.001) (0.001)

District RIGHT MARGIN -0.003 0.015 0.021(0.009) (0.015) (0.016)

District SECOND -0.507 1.149 1.197(0.469) (0.822) (0.862)

Economic SKILL -3.719 -2.773 -2.487 -3.803(3.244) (3.596) (3.680) (4.328)

Control CATHOLICS 0.008 0.005 0.001(0.005) (0.006) (0.006)

Control NON AGRICULTURAL 0.025** 0.024* 0.027*(0.012) (0.013) (0.015)

CONSTANT 1.769 -0.079 -1.367 0.983(1.954) (2.348) (2.523) (2.943)

Regional Fixed Effects � � � ✓R2 0.690 0.713 0.729 0.766CPC 95.07% 94.72% 94.72% 96.48%N 284 284 284 284�� -60.555 -56.082 -52.871 -45.789p(χ2) 0.000 0.000 0.000 0.000BIC 143.706 157.356 162.232 176.313

Standard errors in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01R2: McFadden’s R2; CPC: correctly predicted cases, baseline 55.28%.

We estimate four different model specifications. In all models the outcome is whether a member of

the Reichstag voted yes or no on the proposal to introduce proportional representation. All four

models are presented in Table 2. The first specification, Model 1, is a probit model which includes

variables relating to the three theoretically interesting concepts; the seat-vote disproportionality, the

strength of left parties, and the average skill level in a district. The second specification, Model 2,

adds a number of controls for the tightness of the election in a district (RIGHT MARGIN, SECOND)

as well as the controls (CATHOLICS, NON AGRICULTURAL). The third model also includes two

interactions, one between SOCIALVOTE and SECOND and one between RIGHT MARGIN and

SOCIALVOTE. These interactions reflect the theoretical argument; when politicians are elected in

a second round, their district is already competitive. Hence the strength of the left parties or the

right margin should add less to their propensity to support Proportional Representation than if

they were elected in the first round. Finally, in model 4 we also include regional fixed effects as an

illustration of the robustness of the results.

19

As hypothesized, we find a negative effect for seat-vote disproportionality across all specifica-

tions. The more disadvantageous the seat-vote ratio under majoritarian elections is for a party, the

more likely it will be that its members support a change to proportional representation (Rokkan

II). An increase in the strength of the left parties in a district increases the likelihood that the

politicians will support a change in electoral rules (Rokkan I). Finally, there is no effect for the av-

erage skill level in a district on the willingness of the representative to support changes in electoral

institutions. A comparison of the estimates from Model 2 (without interactions) and Model 3 (with

interactions) shows how the size of the coefficient for SOCIALVOTE more than triples while the

interaction (with SECOND) is negative and significant. The explanation for the increase in the size

of the coefficient for SOCIALVOTE is that we now allow the model to have two different effects for

SOCIALVOTE: one for politicians elected in the first round and one for those politicians elected

in the second round. The left threat exerts an effect for politicians which were elected in the first

round while those representatives which had to face a run-off are already in competitive districts

and the left threat should affect them less.The change of β̂SOCIALVOTE from Model 2 to Model 3 is

consistent with the theoretical predictions of the first Rokkanian hypothesis. Finally, adding fixed

effects does not change the estimates in any meaningful way with the exception that the interaction

between RIGHT MARGIN and SOCIALVOTE achieves statistical significance.4

Figure 1: Support of PR Adoption

0 10 20 30 40 50

0.0

0.2

0.4

0.6

0.8

1.0

Strength of SPD in a District

Pred

icted

Pro

babi

lity

Elected in First RoundElected in Second Round

To fully illustrate these effects we

present a figure with predicted probabil-

ities to clarify the interaction effects (see

Figure 1). We plot the predicted probabil-

ities for two hypothetical cases where the

only difference is that one legislator was

elected in the first round and the other

was elected in a run-off. Simulation re-

sults rest on estimates from Model 3 and

the predicted probabilities are displayed with ± one standard deviation. The marginal effect of the

strength of the SPD in a district has a positive and significant effect. In addition, the magnitude of

effect is stronger in those districts which did not require a run-off election.4A χ2 test whether SECOND and SECOND*SOC.VOTE are significant predictors or not yields a test statistic of

χ22 = 13.28 and a p−value of 0.001 based on Model 3 (for Model 4: χ2

2 = 10.24 and a p−value of 0.005).

20

Finally we turn our attention to the economic predictors. In all specifications we find that the

average skill ratio of a district has no significant influence on the probability of a legislator voting

yes on the adoption of proportional representation. But we do find an effect for economic modern-

ization (NON AGRICULTURE) but this effect is not robust and disappears in alternative model

specifications. The non-finding for SKILL does not depend on the inclusion of NON AGRICUL-

TURE; all models support the null hypothesis that there is no relation between the probability of

a yes vote and the average skill level.5

In addition to these results, we perform a number of robustness checks and present all the

estimation results in the appendix. We present the same models but use two alternative measures

of skilled labor force – skill level of total work force as well as a measure of skill that is aggregated

over regions. The results are presented in Table 5 and Table 7 in the appendix. Among these

alternative models presented in the appendix, we also estimate a model that examines the vote

choice of non-socialist representatives in the Reichstag only and excludes the Social Democratic

candidates. The results reinforce our confidence in the explanatory value of the electoral threat of

the left on the political support for changes in electoral rules (see Table 11 in the appendix). Finally,

we also re-estimate alternative models where we include a measure of population size to control for

possible effects of malapportionment (Table 9 in the appendix). In all specifications the results

remain stable; the seat-vote disproportionality has a significant and negative effect on support for

proportional representation, the strength of the left vote in a district has a positive and significant

effect, and there is no statistical relation between skill levels and voting behavior.

The two Rokkanian arguments do not apply to the same level of analysis. The left threat is

a district-level variable whereas the seat vote disproportionality is a party-level variable. To take

these different levels fully into account empirically, we re-estimate Models 1 through 4 including re-

specifications that take the hierarchical data structure into account (Steenbergen and Jones, 2002;

Gelman and Hill, 2007). Table 3 presents the estimation results for Models 5 through 8 which

are hierarchical probit models, where the first level is the district level and the second level is the

partisan level.

Again, the outcome variable in both models is the vote of a single member of parliament. The

specification is identical to the models in Table 2 with the exception that we model the seat vote5In an estimation of Model 4 that excludes NON AGRICULTURE, we find β̂SKILL = −4.5 and a p−value of 0.289.

21

disproportionality (SEAT VOTE DISPR) as a level 2 variable and include a random effect which

varies over parties. Subscript i describes legislators and subscript j pertains to parties:

Pr(Voteij = 1) = Φ(β0j +βXij)β0j = α0 + α1 ⋅ SVDj + ujuj ∼ N(0,σu)

The first line describes the level 1 equation where Xij denotes the matrix of the explanatory

variables and β is a vector of regression coefficients with the same number of elements as there

are explanatory variables. The substantive effects and statistical significance of the three most

important variables that model the three competing explanations about the adoption of proportional

representation (seat-vote disproportionality, skill level and left strength) remain unchanged when

compared to previous specifications presented in Table 2.

Table 3: Hierarchical Probit Models on PR Vote

Model 5 Model 6 Model 7 Model 8

Partisan SEAT VOTE DISPR. -0.317*** -0.362*** -0.381*** -0.393***(0.102) (0.112) (0.107) (0.106)

District SOCIALVOTE 0.033*** 0.035** 0.087* 0.088*(0.013) (0.015) (0.048) (0.053)

District SECOND*SOC.VOTE -0.072 -0.079(0.045) (0.049)

District RIGHT MARGIN*SOC.VOTE -0.000 -0.001(0.001) (0.001)

District RIGHT MARGIN -0.004 0.006 0.016(0.010) (0.017) (0.018)

District SECOND -0.965 0.475 0.724(0.616) (1.001) (1.048)

Economic SKILL -2.965 -2.774 -2.351 -2.643(3.780) (4.338) (4.403) (5.040)

Control CATHOLICS -0.001 -0.004 -0.008(0.008) (0.008) (0.009)

Control NON AGRICULTURAL 0.021 0.020 0.021(0.013) (0.013) (0.016)

CONSTANT 1.001 0.306 -0.609 0.978(2.299) (2.920) (3.026) (3.366)

ln(σ2j ) 0.007 0.089 -0.070 -0.169

(0.856) (0.851) (0.870) (0.930)Regional Fixed Effects � � � ✓CPC 95.07% 95.07% 95.07% 96.13%N 284 284 284 284�� -50.375 -47.643 -45.906 -40.987p(χ2) 0.000 0.003 0.003 0.007BIC 128.995 146.127 153.951 172.358

Standard errors in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01R2: McFadden’s R2; CPC: correctly predicted cases, baseline 55.28%.

22

Moreover, the effects of these variables do not change across all models presented in Table 3.

In these specifications, the measure of economic development (NON AGRICULTURAL) no longer

reaches statistical significance at conventional levels, when compared to Table 2. The coefficients of

the interaction between the second round and the social democratic vote and the right margin and

the social democratic vote also do not reach statistical significance. These changes notwithstanding,

we find that the tests of the relevant theoretical hypotheses remain unchanged. The skill level in

a district has no effect on the probability that a legislator in the Reichstag will support a change

of the electoral system. By contrast, political variables exercise a considerable effect. Increases

in seat-vote disproportionality and the strength of the left-wing candidate competing in a district

increase the probability of support for proportional representation.

Figure 2 presents predicted probabilities for support of changes in the electoral system. The

purple curve presents changes in the probability of support for proportional representation of a

legislator representing the National Liberal party for different levels of electoral strength of the Social

Democratic challenger (left threat). The blue curve represents similar probabilities for members of

the Zentrum.

Figure 2: Seat Vote Disproportionality

0 10 20 30 40 50

0.0

0.2

0.4

0.6

0.8

1.0

Strength of SPD in a District

Pred

icted

Pro

babi

lity National Liberals

Catholics (Zentrum)

National Liberal legislators are

more likely to support a change in the

electoral system even if there is no left

threat at all. This can be explained by

the unfavorable allocation of seats to

votes experienced by this party during

the 1912 election. By contrast, a hy-

pothetical legislator representing the

Zentrum found the existing electoral

rules more attractive. The Zentrum, a party with strong regional strongholds in Bavaria, Baden and

the Ruhr benefited from allocation of votes to seats under the existing electoral rules throughout

the Imperial Period. During the 1912 elections, the Zentrum ended up receiving a surplus in the

allocation of seats of (+6.1%-points). Despite these differences across parties, the plot reveals that

an increase in the left wing threat increased the probability of support of changes in electoral rules

for both the National Liberal and the Zentrum representatives.

23

One might object to the operationalization of skill level as a district variable and claim that the

skill mechanism rather works through the party. In that logic, the individual representative might

be responsive even if his own district is not endowed with a high skill level but rather most other

members of his party have districts with above average skill levels. We also provide estimation results

for that operationalization where we first aggregate average skill level for each party separately to

the party level (average of skill level for all districts held by a party) and add this variable as a

party-level variable to the hierarchical model.The results are substantively identical to the models

presented in Table 3. There is a negative and significant effect for seat-vote disproportionality, a

positive and significant effect for the strength of the left in a district and no relationship between the

skill level and the propensity to support proportional representation (see Table 10 in the appendix).6

4.2 Party Defection and Left Threat

The two political variables that impinge upon the political decisions of legislators – the left threat

in individual districts vs. party interests – reinforce each other on some occasions, but can, on other

occasions, pull in different directions. In this section, we examine those cases where legislators

defected from their party position and voted against the party’s interest. Can such defections be

explained by individual-district motivations?

To illustrate how district-level and partisan consideration can come into conflict with each

other, consider the case of politicians, from the two parties opposing the adoption of proportional

representations – Zentrum and Conservatives – , who face a strong social democratic threat in

their district. In this case, partisan considerations pull the politician towards support for existing

electoral rules, while district-level calculations pull in the direction of changes in electoral systems.

The defection of these politicians from the position advocated by their party can be explained by

district level competition. This is the case, for example of Eugen Schatz – a Zentrum politician

from Saargemünd-Forbach who supported the adoption of proportional representation. Schatz had

won his seat with a narrow margin against a Social Democratic opponent. Albert Thumann, a

politician from Gebweiler and another Zentrum politician who defected from the position of his

party – had also clinched victory narrowly (with a majority of 55 percent of the votes) against a

Social Democratic challenger.6We also estimated a model in which we let the coefficients vary over parties and uncover the same estimates with

respect to significance level and direction of effect.

24

Table 4: Logit Models on Party Defection

Model 9 Model 10 Model 11 Model 12 Model 13

SOCIALVOTE 0.092** 0.118** 0.110**(0.040) (0.052) (0.052)

SKILL -18.703 -15.080 -12.926(12.452) (13.562) (15.403)

RIGHT MARGIN 0.006 -0.018 0.012(0.026) (0.027) (0.028)

NON AGRICULTURAL 0.002 0.018 -0.009(0.032) (0.030) (0.036)

CATHOLICS -0.058 -0.025 -0.067(0.043) (0.044) (0.045)

SECOND -3.116 -2.121 -2.747(1.976) (1.769) (1.980)

CONSERVATIVE -2.911* -1.840 -5.336 -2.759 -5.983(1.599) (1.482) (3.549) (3.454) (3.922)

Constant -3.754*** 8.769 0.554 8.811 9.473(0.883) (7.251) (3.522) (10.099) (11.487)

N 108 108 108 108 108�� -14.871 -21.940 -2.401 -9.802 -4.560p(χ2) 0.040 0.146 0.202 0.500 0.323BIC 43.789 57.927 37.577 52.378 46.577Standard errors in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01 – Logit modelwith penalized maximum likelihood function, “firthlogit” (Firth, 1993).

In Table 4 we present an array of five different models explaining defection from the party line

for the Catholics (Zentrum) and the Conservatives. As hypothesized, the strength of the vote of

the Social Democratic candidate in a district (SOCIALVOTE) can explain party defection. All

specifications include the alternative explanatory variable proposed by Cusack et al. (2007, 2010)

which is the district skill level (SKILL) as well as the control variables from Models 1-8. We also

add a dichotomous indicator for Conservatives to allow for party differences. In the appendix, we

present the same models estimated only on members of Zentrum as well as with an alternative

underlying model (see Table 12, Table 13, and Table 14). Since the variable (SECOND) causes

perfect separation (it perfectly predicts the outcome) (Zorn, 2005), we use a logit model based on

a penalized maximum likelihood function to overcome this problem (Firth, 1993) which is similar

to a Bayesian estimation of the model.

Regardless which model specification, all results in Models 9 through 13 (and the alternative

specifications in the appendix) support the first Rokannian hypothesis that legislators facing a strong

raising left, support proportional representation. Those members of Zentrum and the Conservatives

25

who defected from the party line were facing considerably stronger left candidates in their districts.7

Those members of Zentrum who voted against their party line faced significantly stronger left

contenders in their districts. This illustrates the strength of Rokkan’s two hypotheses when they are

allowed to jointly explain the same observations on the party level as well as the individual district

level. We have now not only established that the two Rokkanian hypotheses can explain the overall

vote outcome and the adoption of PR, but we also illustrated that Rokkan I (Social Democratic

vote share as threat) explains party defection among those parties that share an electoral advantage

from using SMD and thereby shows that these two explanations are not two different roads leading

to PR but arguments which apply simultaneously at different levels of analysis within the same

case.

5 Conclusion

We have started this paper by noting a paradoxical situation in the recent literature examining the

origin of proportional representation. The flurry of papers in prominent publications notwithstand-

ing, one was left wondering which of explanations does a better job in accounting for the choice

of electoral rules in European countries during the first decades of the 20th century. The current

debate has been framed as a horse-race between the first Rokkanian hypothesis and the economic

arguments formulated by Cusack et al. (2007), to the neglect of alternative explanations (which

are also Rokkanian in origin) that attribute an important role to partisan calculations (exception is

Calvo 2009).

In this paper, we have argued that political demand for the adoption of proportional represen-

tation can be attributed to factors that operate at two distinct levels. One set of considerations

operate at the level of the party. Disproportionalities in the translation of votes to seats and per-

ceptions of possible gains in the allocation of seats under the new electoral rules affect partisan

considerations about the desirability of changes in electoral rules. These factors explain variation in

the partisan positions among parties on the right and the position of social democratic parties on

the issue of electoral reforms. A second set of factors that impinge upon the decisions of politicians

operate at the level of the electoral district. Here vulnerability to electoral competition from left7Testing the difference in means of SPD vote share, we find a test statistic of t = 2.60 and a p−value of 0.013.We

also used a non-parametric test, randomization inference, which yields the same result. The average vote share ofSPD in the districts of the six deviators is significantly higher than the average SPD vote share in any random draw ofsix members of the Zentrum. We used 100,000 randomization and found a p−value of 0.02 (Gerber and Green, 2012).

26

wing candidates can explain some additional variation in political support for changes in electoral

rules among candidates on the political right and the willingness of the latter to support or deviate

from the position advocated by their parties. While these political hypotheses go back to the work

of Rokkan, we reformulate and clarify the Rokkanian conjectures. We argue that it is more use-

ful to think about the two political hypotheses about the adoption of proportional representation

as identifying complementary constraints on politicians, rather than as ’separate paths’ chosen by

different countries. Empirically, we demonstrate that both logics are at work within the same case

and that they identify different sources of demand for changes in electoral rules.

Our paper also proposes a new empirical strategy to examine the question of changes in electoral

rules that departs from a reliance on a cross-national analysis. We propose to change the unit of

analysis from countries to individual legislators and bring back individual politicians to the center

of the analysis. This empirical strategy has several advantages over cross-national research. The

first is that it allows us to test existing explanations at the salient level, the level of the district,

rather than the level of the country. Secondly, it allows to consider all the relevant political and

economic hypotheses and not just a subset of the latter. Finally, we improve upon existing studies

by developing more precise measures of the salient political and economic variables.

We find that political support for the adoption of proportional representation was affected by

partisan considerations about the disproportionality in the translation of seats to votes and by po-

litical conditions in a district. We do not find that legislative support can be explained by skill levels

regardless of the many alternative specifications we employ. Our empirical analysis disconfirms the

economic hypothesis about the adoption of proportional representation. But given that Germany

is a ‘hard case’ for Rokkanian explanations, we believe that our findings weaken significantly our

confidence in the comparative explanatory power of economic hypotheses for the adoption of pro-

portional representation. Our findings also qualify the Rokkanian hypothesis about the importance

of the Social Democratic threat as a determinant of the adoption of proportional representation in

Germany. We show that political choices made by German legislators over the adoption of pro-

portional representation can be explained by a combination of partisan dissatisfaction and district

level vulnerabilities. Contra Rokkan, we show that choices over electoral rules are affected by a

combination of partisan and electoral considerations and not by district level vulnerabilities to the

Social Democratic threat only.

27

References

Ahmed, Amel. 2010. “Reading history forward: The origins of electoral systems in European democracies.”Comparative Political Studies 43(8-9):1059–1088.

Alesina, Alberto, Edward Glaeser and Bruce Sacerdote. 2001. Why Doesn’t the US Have a European-StyleWelfare System? Technical report National Bureau of Economic Research.

Anderson, Margaret Lavinia. 2000. Practicing Democracy: Elections and Political Culture in Imperial Ger-many. Princeton: Princeton University Press.

Andrews, Josephine T. and Robert W. Jackman. 2005. “Strategic fools: electoral rule choice under extremeuncertainty.” Electoral Studies 24(1):65–84.

Bade, Klaus. 1980. Massenwanderung und Arbeitsmarkt im deutschen Nordosten 1880- 1914. Archiv fürSozialgeschichte.

Benoit, Kenneth. 2004. “Models of electoral system change.” Electoral studies 23(3):363–389.

Bertram, Juergen. 1964. Die Wahlen zum deutschen Reichstag vom Jahre 1912: Parteien und Verbände inder Innenpolitik des wilhelminischen Reiches. Droste.

Blais, André, Agnieszka Dobrzynska and Indridi Indridason. 2005. “To adopt or not to adopt proportionalrepresentation: the politics of institutional choice.” British Journal of Political Science pp. 182–190.

Boix, Carles. 1999. “Setting the Rules of the Game: The Choice of Electoral Systems in Advanced Democ-racies.” The American Political Science Review 93(3):609–624.

Boix, Carles. 2010. “Electoral markets, party strategies, and proportional representation.” American PoliticalScience Review 104(2):404–413.

Brandenburg, Erich. 1917. 50 Jahre Nationalliberale Partei: 1867-1917. Schriftenvertriebsstelle der Nation-alliberalen Partei Deutschlands, Kalkoff.

Braunias, Karl. 1932. Das parlamentarische Wahlrecht: ein Handbuch über die Bildung der gesetzgebendenKörperschaften in Europa. Vol. 2 W. de Gruyter & co.

Bundesarchiv, Berlin Lichterfeld. 1912. Betrachtungen über die Einteilung der Reichstagswahlkreise. R1501/309-363.

Calvo, Ernesto. 2009. “The Competitive Road to Proportional Representation. Partisan biases and electoralregime change under increasing party competition.” World Politics 61(2):254–95.

Colomer, Josep M. 2005. “It’s parties that choose electoral systems (or, Duverger’s laws upside down).”Political Studies 53(1):1–21.

Cox, Gary W. 1997. Making Votes Count: Strategic Coordination in the World’s Electoral Systems. Cam-bridge: Cambridge University Press.

Cusack, Thomas, Torben Iversen and David Soskice. 2007. “Economic Interests and the Origins of ElectoralSystems.” The American Political Science Review 101(3):373–391.

Cusack, Thomas, Torben Iversen and David Soskice. 2010. “Coevolution of Capitalism and Political Repre-sentation: The Choice of Electoral Systems.” The American Political Science Review 104(2):393–403.

28

Duverger, Maurice. 1954. Political parties: Their organization and activity in the modern state. New York:John Wiley.

Fairbairn, Brett. 1990. “Authority vs. democracy: Prussian Officials in the German Elections of 1898 and1903.” The Historical Journal 33(4):811– 838.

Firth, David. 1993. “Bias Reduction of Maximum Likelihood Estimates.” Biometrika 80(1):27–38.

Fremdling, Rainer and Richard Tilly. 1979. Industrialisierung und Raum. Studien zur regionalen Differen-zierung im Deutschland des 19.Jahrhunderts. Stuttgart: Klett-Cotta.

Fricke, Dieter. 1961. “Der deutsche Imperialismus und die Reichstagswahlen von 1907.” Zeitschrift fürGeschichstwissenschaft 9(3):538– 576.

Gagel, Walter. 1958. Die Wahlrechtsfrage in der Geschichte der deutschen liberalen Parteien: 1848-1918.Vol. 12 Droste Verlag.

Gelman, Andrew and Jennifer Hill. 2007. Data Analysis Using Regression and Multilevel/Hierarchical Models.Cambridge: Cambridge University Press.

George, Alexander L. and Andrew Bennett. 2005. Case Studies and Theory Development in the SocialSciences. Cambridge: MIT Press.

Gerber, Alan S. and Donald P. Green. 2012. Field Experiments: Design, Analysis, and Interpretation. NewYork: W.W. Norton.

Hatschek, Julius. 1920. Kommentar zum Wahlgesetz und zur Wahlordnung im Deutschen Kaiserreich. Berlinund Leipzig: Walter de Gruyter.

Hohls, Rüdiger and Hartmut Kaelble. 1989. Die regionale Erwerbsstruktur im deutschen Reich und in derBundesrepublik Deutschland. St. Katharinen : Scripta Mercaturae Verlag.

Huber, Evelyne, Charles Ragin and John D Stephens. 1993. “Social democracy, Christian democracy, con-stitutional structure, and the welfare state.” American Journal of Sociology pp. 711–749.

Kaiserliches-Statistisches-Amt. 1898. Statistik des Deutschen Reichs. [Band 112]. Berlin: Verlag desKoniglich Preussischen Statistischen Bureaus.

Kaiserliches-Statistisches-Amt. 1899a. Die berufliche und soziale Gliederung des deutschen Volkes nach derBerufszählung vom 14. Juni 1895. Berlin: Verlag des Koniglich Preussischen Statistischen Bureaus.

Kaiserliches-Statistisches-Amt. 1899b. Statistik des Deutschen Reichs. Volumes 117 and 118. Berlin: VonPuttkammer und Mühlbrecht.

Kaiserliches-Statistisches-Amt. 1907. Gewerbliche Betriebsstatistik. Berlin: Von Puttkammer undMühlbrecht.

Kaiserliches-Statistisches-Amt. 1909. Statistik des Deutschen Reiches, Neue Folge, Volumes 218 and 219.Berlin: Von Puttkammer und Mühlbrecht.

Kaiserliches-Statistisches-Amt. 1913. Die Reichstagswahlen von 1912. In Statistik des Deutschen Reiches,Band 250, Zweites Heft. Berlin: Von Puttkammer und Mühlbrecht.

29

Kreuzer, Marcus. 2010. “Historical Knowledge and Quantitative Analysis: The Case of the Origins ofProportional Representation.” The American Political Science Review 104(2):369–392.

Kühne, Thomas. 1994. Landtagswahlen zwischen korporativer Tradition und politischem Massenmarkt. InDreiklassenwahlrecht und Wahlkultur in Preussen 1867-1914. Düsseldorf : Droste.

Kuo, Alexander. 2010. The political origin of firms’ strategies. PhD thesis Department of Political Science,Stanford University.

Lijphart, Arend. 1990. “The political consequences of electoral laws, 1945-85.” The American Political ScienceReview pp. 481–496.

Luebbert, Gregory M. 1991. Liberalism, fascism, or social democracy: Social classes and the political originsof regimes in interwar Europe. Oxford University Press, USA.

Penadés, Alberto. 2008. “The Institutional Preferences or Early Socialist Parties: Choosing Rules for Gov-ernment.” Controlling Governments – Voters, Institutions, and Accountability José María Maravall.

Penadés, Alberto. 2011. “Electoral Reform in Early Democracies: The Right Wing.” Presented at the Con-ference on Historical Political and Economic Development in Western Europe: Recent Advances inComparative Politics, Madrid .

Person, Torsten and Guido Tabellini. 2004. “Constitutions and economic policy.” The Journal of EconomicPerspectives 18(1):75–98.

Persson, Torsten and Guido E. Tabellini. 2005. The economic effects of constitutions. The MIT Press.

Reibel, Carl-Wilhelm. 2007. Handbuch der Reichstagswahlen 1890- 1918. Bundnisse, Ergebnisse, Kandi-daten. Düsseldorf: Droste.

Renwick, Alan. 2010. The Politics of Electoral Reform: Changing the Rules of Democracy. CambridgeUniversity Press.

Rogowski, Ronald. 1987. “Trade and the variety of democratic institutions.” International Organization41(02):203–223.

Rokkan, Stein. 1970. Citizens Elections Parties. New York: David McKay Co.

Schmädeke, Jurgen. 1995. Wählerbewegung im Wilhelminischen Deutschland: die Reichstagswahlen von 1890bis 1912. Berlin: Akademie Verlag.

Simon, Klaus. 1969. Die württembergischen Demokraten: ihre Stellung und Arbeit im Parteien- und Verfas-sungssystem in Wurttemberg und im Deutschen Reich 1890-1920. Vol. 52 Kohlhammer.

Steenbergen, Marco R. and Bradford S. Jones. 2002. “Modeling Multilevel Data Structures.” AmericanJournal of Political Science 46:218–237.

Thelen, Kathleen. 2004. How Institutions Evolve: The Political Economy of Skills in Germany, Britain, theUnited States and Japan. New York: Cambridge University Press.

Zorn, Christoper. 2005. “A Solution to Separation in Binary Response Models.” Political Analysis 13(2):157–170.

30

Appendix

Appendix A1: How Vote Abstention Recommendations led to Social Democratic

Victories p. 32

Appendix A2: A Measure of Skill Level p. 32

Appendix A3: Alternative Skill Measures p. 33

A3.1: Skill Level of Entire Labor Force p. 33

A3.2: Skill Level as a Regional Measure p. 34

Appendix A4: Effect of District Population Size p. 36

Appendix A5: Skills as a Party Level Variable p. 38

Appendix A6: Model Estimation on Non-Left Politicians p. 38

Appendix A7: Party Defection p. 40

Appendix A8: Skilled and Unskilled workers in 1895 p. 42

31

Appendix A1: How Vote Abstention Recommendations led to Social Democratic

Victories

To illustrate how recommendations to their voters for abstention made by parties on the rightcontributed to the victory of the Social Democratic candidate, consider the following examples(from the 1912 election): In Potsdam-Osthavelland Spandau, the candidates in the runoff were arepresentative of the Imperial Party and a Social Democratic candidates. The Free Liberal Party(‘Fortschrittliche Volkpartei’), which had received 24 percent of the votes in the first round, recom-mended their voters to abstain, which led to the victory of the Social Democratic candidate. (Reibel2007: 163). In Heiligenstadt Worbis, a Free Liberal candidate encountered a Socialist in runoffs.In this case, an electoral committee of the ‘united conservative parties’ (‘Vereinigte RechtstehendeParteien’), which included National Liberals, Conservatives and the Economic Union recommendedtheir voters abstention during runoffs. This led to the victory of the Social Democratic candidate,Cohn (Reibel 2007: 530). This within-country variation in the electoral strategies during runoffsis conditioned by distinct distributions of voter preferences across districts, by previous patterns ofelectoral competition and so on. One specific form of failed coordination occurs when a right partyrecommends abstention in the second round and supports neither of the candidates. This is mostsevere if one of the two remaining candidates belongs to the Social Democrats. During the 1912election, we can identify at least 13 districts in which a non-socialist party recommended absten-tion and the Social Democrats eventually won the district in the run-off. While runoffs offer anopportunity to coordinate, due to conflicts within the right the presence of runoffs is not sufficientto assure coordination.

Appendix A2: A Measure of Skill Level

The data on skill distributions has the following structure. Every occupation is classified first ac-cording to agriculture, mining and producing industry, or trade and commerce. Within these broadcategories there are two further levels. To illustrate this, consider the occupational classificationof goldsmiths. This person makes jewelry and therefore (s)he is in the second sector (mining andproducing industry), within that sector (s)he belongs to the vocational group “V” (metalworkingclass) . If the person does not use copper but gold (s)he is in the vocational group “V” and a type“a” (metalworking with gold or silver). Germany’s 1895 occupational census collected informationon the ratio of skilled to unskilled workers at this very low level of occupational aggregation. Weknow exactly how many people employed in “Va” were skilled.

Unfortunately, as part of the 1905 census, the German Statistical Office did not collect similardata measuring the ratio of skilled to unskilled workers for each occupation. The only informationthat exists in the 1905 census is a comparative table on the number of apprentices (Lehrlinge) foreach occupation, which compares the 1895 and 1905 census. Given that this table reveals that theovertime change in these ratios is very small, one can assume that the skill ratio remains constantover time for each occupation. What changes, however, is the occupational make-up of a district.This requires us to recalculate the skill ratio using the information collected as part of the 1905

32

census, the last census of the period. We use the same procedure as in the calculation of the skillratios for 1895, using this time the information on the distribution of voters across localities thatis presented in the 1905 Gewerbezählung (Statistik des Deutschen Reiches 1909: volumes 218 and219). Again, we construct two measures that measure the share of skilled workers to industrialworkers only and the share of skilled workers to the entire labor force.

Appendix A3: Alternative Skill Measures

Here we present the estimates of models 1 through 8 (Table 2 and Table 3) for a number of alternativemeasures for the skill variable. Instead of relying on a measure of the skill of the industrial workforce,we use a measure of the skill ratio in the total workforce (SKILL ALL). In a second step we useregional average values instead of district measures, and finally, we will aggregate it over partydistricts and compute a party-level variable, similar to seat-vote disproportionality. We believe thatthe best specifications are the ones we present in the main body of the paper, but use this appendixto present alternative specifications to increase the confidence in our main results. All results inthis appendix confirm our substantive claims. Changes in operationalization do not lead to changesin the magnitude or the statistical significance of our estimates. Table 5 replicates the results inTable 2 using the alternative skill measure, while Table 6 replicates the results of Table 3.

A3.1: Skill Level of Entire Labor Force

We first perform a robustness check for an alternative measure of skill level. Instead of relying on theshare of skilled industrial labor force, we use the average skill level over the entire labor population.We present here replications of all eight models in Table 2 and Table 3 and will name the models1a through 8a. The results are presented in Table 5 and Table 6. The substantive results remainunchanged; seat-vote disproportionality and the left threat are significant explanatory factors ofsupport for adoption of proportional representation, whereas the coeffient for the skill level has nosignificant effect.

33

Table 5: Replication of Table 2 with Alternative Skill Measure

Model 1a Model 2a Model 3a Model 4aPartisan SEAT VOTE DISPR. -0.316*** -0.391*** -0.407*** -0.414***

(0.040) (0.057) (0.058) (0.061)District SOCIALVOTE 0.020* 0.026** 0.110** 0.099**

(0.011) (0.013) (0.044) (0.047)District SECOND*SOC.VOTE -0.094** -0.092**

(0.039) (0.042)District RIGHT MARGIN*SOC.VOTE -0.001 -0.001*

(0.001) (0.001)District RIGHT MARGIN -0.003 0.014 0.020

(0.009) (0.015) (0.016)District SECOND -0.493 1.158 1.192

(0.467) (0.822) (0.859)Economic SKILL ALL 0.030 -0.037 -0.027 -0.048

(0.019) (0.057) (0.058) (0.068)Control CATHOLICS 0.008 0.005 0.001

(0.005) (0.006) (0.006)Control NON AGRICULTURAL 0.047 0.040 0.055

(0.035) (0.035) (0.041)CONSTANT -1.342** -1.735** -2.863*** -1.268

(0.635) (0.727) (0.991) (1.137)Regional Fixed Effects � � � ✓R2 0.693 0.712 0.729 0.765CPC 94.72% 94.72% 94.72% 96.48%N 284 284 284 284�� -59.870 -56.170 -52.989 -45.924p(χ2) 0.000 0.000 0.000 0.000BIC 142.336 157.532 162.468 176.582

* p < 0.10, ** p < 0.05, *** p < 0.01, McFadden’s R2, E(y) = 0.553Table 6: Replication of Table 3 with Alternative Skill Measure

Model 5a Model 6a Model 7a Model 8aPartisan SEAT VOTE DISPR. -0.328*** -0.362*** -0.382*** -0.393***

(0.104) (0.112) (0.107) (0.106)District SOCIALVOTE 0.030** 0.035** 0.086* 0.087*

(0.013) (0.015) (0.048) (0.052)District SECOND*SOC.VOTE -0.072 -0.078

(0.045) (0.049)District RIGHT MARGIN*SOC.VOTE -0.000 -0.001

(0.001) (0.001)District RIGHT MARGIN -0.004 0.006 0.015

(0.010) (0.017) (0.018)District SECOND -0.951 0.483 0.723

(0.613) (0.999) (1.045)Economic SKILL ALL 0.027 -0.033 -0.023 -0.033

(0.020) (0.069) (0.070) (0.079)Control CATHOLICS -0.000 -0.003 -0.007

(0.008) (0.008) (0.009)Control NON AGRICULTURAL 0.041 0.034 0.040

(0.040) (0.040) (0.047)CONSTANT -1.619* -1.353 -2.026* -0.565

(0.846) (1.004) (1.185) (1.337)ln(σ2

j ) 0.030 0.086 -0.070 -0.162(0.855) (0.851) (0.869) (0.927)

Regional Fixed Effects � � � ✓CPC 94.36% 95.07% 94.71% 96.13%N 284 284 284 284�� -49.723 -47.732 -45.992 -41.038p(χ2) 0.000 0.003 0.003 0.007BIC 127.691 146.304 154.122 172.460

* p < 0.10, ** p < 0.05, *** p < 0.01, McFadden’s R2, E(y) = 0.553A3.2: Skill Level as a Regional Measure

So far we have used measures of skills at the level of the electoral district and measured either theshare of skilled workers in the industrial labor force or in the entire labor force. In a next step we

34

will use a skill measure which aggregate skills to the level of the region. The economic argumentfor the adoption of proportional representation, advanced by Cusack Iversen and Sokice can beunderstood as an argument about regional political economies. In this interpretation, politiciansdo not respond to the skill composition of a district but rather of the larger political economy inwhich they are embedded. To test this proposition, we derive a final measure of skill level (SKILLREGION) where we use the average value for each of the 29 political units of Imperial Germany.We name all models 1b through 8b analogous to section A3.1.

Table 7: Replication of Table 2 with Regional Skill Measure

Model 1b Model 2b Model 3b Model 4b

Partisan SEAT VOTE DISPR. -0.314*** -0.405*** -0.417*** -0.427***(0.040) (0.058) (0.059) (0.063)

District SOCIALVOTE 0.028*** 0.028** 0.113** 0.098**(0.011) (0.013) (0.044) (0.049)

District SECOND*SOC.VOTE -0.091** -0.084*(0.040) (0.044)

District RIGHT MARGIN*SOC.VOTE -0.001 -0.001*(0.001) (0.001)

District RIGHT MARGIN -0.003 0.015 0.022(0.009) (0.015) (0.016)

District SECOND -0.470 1.109 1.130(0.470) (0.827) (0.889)

Economic SKILL REGION -15.590* -16.207* -14.245 -33.747**(8.366) (8.865) (9.203) (14.134)

Control CATHOLICS 0.009 0.006 0.001(0.005) (0.006) (0.006)

Control NON AGRICULTURAL 0.026** 0.024* 0.031**(0.013) (0.013) (0.015)

CONSTANT 8.749* 7.747 5.456 18.328**(4.939) (5.242) (5.472) (8.377)

Regional Fixed Effects � � � ✓R2 0.695 0.720 0.734 0.774CPC 95.07% 95.07% 94.72% 96.83%N 284 284 284 284�� -59.505 -54.756 -51.928 -44.104p(χ2) 0.000 0.000 0.000 0.000BIC 141.606 154.704 160.345 172.943

* p < 0.10, ** p < 0.05, *** p < 0.01, McFadden’s R2, E(y) = 0.553As in the above replication, there is no model in which the skill level – however measured – is

positively related to supporting a change of the electoral system. When using the regional measurethe relationship is actually negative and significant. Hence, these results here outright refute theclaim of Cusack et al. (2007). Again, we present the best operationalization and specification in themain part of the paper and add these models only for the critical reader.

35

Table 8: Replication of Table 3 with Regional Skill Measure

Model 5b Model 6b Model 7b Model 8bPartisan SEAT VOTE DISPR. -0.357*** -0.389*** -0.398*** -0.407***

(0.129) (0.131) (0.125) (0.123)District SOCIALVOTE 0.037*** 0.038** 0.083* 0.080

(0.013) (0.015) (0.048) (0.055)District SECOND*SOC.VOTE -0.058 -0.062

(0.046) (0.052)District RIGHT MARGIN*SOC.VOTE -0.000 -0.001

(0.001) (0.001)District RIGHT MARGIN -0.005 0.006 0.016

(0.010) (0.017) (0.018)District SECOND -0.930 0.254 0.474

(0.626) (1.046) (1.130)Economic SKILL REGION -32.006*** -30.358** -27.442** -47.295***

(12.370) (12.582) (13.423) (17.957)Control CATHOLICS -0.001 -0.003 -0.009

(0.007) (0.008) (0.009)Control NON AGRICULTURAL 0.020 0.019 0.024

(0.014) (0.014) (0.016)CONSTANT 18.130** 16.670** 14.303* 26.990**

(7.328) (7.559) (8.069) (10.670)ln(σ2

j ) 0.481 0.439 0.306 0.214(0.879) (0.866) (0.899) (0.991)

Regional Fixed Effects � � � ✓CPC 93.66% 92.96% 93.31% 96.48%N 284 284 284 284�� -46.875 -44.684 -43.730 -38.343p(χ2) 0.000 0.003 0.006 0.010BIC 121.995 140.209 149.598 167.070

* p < 0.10, ** p < 0.05, *** p < 0.01, McFadden’s R2, E(y) = 0.553

In Table 5, Table 6, Table 7, and Table 8, we present the results of the models with alternativemeasures of skill levels. We find again support for the two driving factors of legislators’ decisions;the partisan SEAT VOTE DISPR as well as the strength of the Social Democrats in a district. Onlyin one models the direct effect of SOCIALVOTE loses statistical significance, but the direction ofthe effect and the size of the coefficient are almost unchanged in comparison to other models. Inthese models we do find a significant effect of SKILL REGION. But this effect is in the oppositedirection of the theoretical prediction. The higher the average skill level is in a region, the lower thepredicted probability for a legislator to support the adoption of proportional representation. Thesemodel illustrate the robustness of our empirical findings.

Appendix A4: Effect of District Population Size

Here we present the estimates of models 1 through 8 and add population size as an explanatoryvariable. These results serve the purpose of illustrating the robustness of our results to possibleconsiderations about malapportionment. The models are labeled 1c to 8c and display the samesubstantive estimates as all the other models; the larger the seat-vote disproportionality is thelower is the probability that a member of the Reichstag will support a change of the electoralsystem. The stronger the SPD vote is in a district, the higher is the probability that a legislator willsupport the introduction of proportional representation. Finally, once we control for a number offactors, we find that population size has an effect only in some model specifications and that effectis negative.

36

Tabl

e9:

Rep

licat

ion

ofTa

bles

2&

3w

ith

Mal

appo

rtio

nmen

t

Mod

el1c

Mod

el2c

Mod

el3c

Mod

el4c

Mod

el5c

Mod

el6c

Mod

el7c

Mod

el8c

Par

tisa

nSE

AT

VO

TE

DIS

PR

.-0

.303

***

-0.4

12**

*-0

.423

***

-0.4

21**

*-0

.321

***

-0.3

74**

*-0

.391

***

-0.3

95**

*(0

.038

)(0

.059

)(0

.060

)(0

.062

)(0

.103

)(0

.113

)(0

.108

)(0

.106

)

Dis

tric

tSO

CIA

LVO

TE

0.02

7**

0.02

9**

0.10

5**

0.09

9**

0.03

6***

0.03

8**

0.08

5*0.

087*

(0.0

11)

(0.0

13)

(0.0

45)

(0.0

48)

(0.0

13)

(0.0

15)

(0.0

46)

(0.0

53)

Dis

tric

tSE

CO

ND

*SO

C.V

OT

E-0

.085

**-0

.090

**-0

.066

-0.0

76(0

.040

)(0

.043

)(0

.043

)(0

.049

)

Dis

tric

tR

IGH

TM

AR

GIN

*SO

C.V

OT

E-0

.001

-0.0

01-0

.000

-0.0

01(0

.001

)(0

.001

)(0

.001

)(0

.001

)

Dis

tric

tR

IGH

TM

AR

GIN

-0.0

060.

009

0.01

8-0

.007

0.00

20.

013

(0.0

09)

(0.0

15)

(0.0

16)

(0.0

10)

(0.0

17)

(0.0

19)

Dis

tric

tSE

CO

ND

-0.6

640.

872

1.07

9-1

.072

*0.

301

0.65

0(0

.482

)(0

.836

)(0

.876

)(0

.625

)(0

.988

)(1

.061

)

Eco

nom

icSK

ILL

-4.4

75-4

.129

-4.1

61-4

.355

-4.0

44-3

.868

-3.6

41-2

.931

(3.3

51)

(3.7

19)

(3.8

48)

(4.3

83)

(3.9

08)

(4.4

24)

(4.5

09)

(5.0

69)

Con

trol

CAT

HO

LIC

S0.

010*

0.00

80.

003

0.00

20.

000

-0.0

06(0

.006

)(0

.006

)(0

.007

)(0

.008

)(0

.008

)(0

.010

)

Con

trol

NO

NA

GR

ICU

LTU

RA

L0.

030*

*0.

030*

*0.

029*

0.02

7**

0.02

8*0.

023

(0.0

13)

(0.0

13)

(0.0

15)

(0.0

13)

(0.0

14)

(0.0

17)

Mal

appo

rtio

nmen

tP

OP

ULAT

ION

-1.1

87-2

.972

**-3

.594

*-1

.712

-1.3

97-2

.408

*-2

.919

-1.0

12(1

.332

)(1

.250

)(1

.878

)(2

.251

)(1

.287

)(1

.331

)(1

.816

)(2

.355

)

CO

NST

AN

T2.

363

0.88

8-0

.135

1.33

71.

801

1.00

10.

159

1.09

8(2

.064

)(2

.444

)(2

.629

)(2

.962

)(2

.411

)(2

.976

)(3

.071

)(3

.361

)ln(σ2 j)

0.01

90.

072

-0.0

87-0

.188

(0.8

56)

(0.8

62)

(0.8

79)

(0.9

35)

Reg

iona

lFix

edE

ffect

s�

��

✓�

��

✓CPC

94.7

2%94

.72%

95.0

7%96

.48%

94.7

2%94

.72%

95.4

2%96

.13%

N28

428

428

428

428

428

428

428

4��

-60.

224

-54.

180

-51.

324

-45.

527

-49.

896

-46.

480

-44.

876

-40.

899

p(χ2)

0.00

00.

000

0.00

00.

000

0.00

00.

004

0.00

30.

010

BIC

148.

693

159.

200

164.

786

181.

437

133.

686

149.

451

157.

540

177.

830

Stan

dard

erro

rsin

pare

nthe

ses;

*p<0.

10,*

*p<0.

05,*

**p<0.

01–R

2:

McF

adde

n’sR

2;C

PC

:cor

rect

lypr

edic

ted

case

s,ba

selin

e55

.28%

.

37

Appendix A5: Skills as a Party Level Variable

We present the estimation results for the hierarchical models with yet another measure of skills.One could presumably argue that the average skill level in a district is not the correct test of theeconomic explanation of proportional representation, but rather whether a party holds a lot of seatsfrom districts with higher skill level or not. To that end we have generated a variable which capturesthe average skill level across all districts won by the respective party.

Pr(Voteij = 1) = Φ(β0j +βXij)β0j = α0 + α1 ⋅ SVDj + α2 ⋅ Skillj + uj

uj ∼ N(0,σu)

We estimate a number of hierarchical models in which the two second level variables are SEATVOTE DISPR. and SKILL PARTY LEVEL. The substantive results are identical to the ones pre-sented in Table 3.

Table 10: Replication of Table 3 with skill measured at the level of the party

Model 5d Model 6d Model 7d Model 8d

Partisan SEAT VOTE DISPR. -0.291** -0.351*** -0.358*** -0.365***(0.119) (0.134) (0.124) (0.122)

District SOCIALVOTE 0.034*** 0.036** 0.088* 0.089*(0.013) (0.015) (0.048) (0.053)

District SECOND*SOC.VOTE -0.074* -0.081(0.045) (0.050)

District RIGHT MARGIN*SOC.VOTE -0.000 -0.001(0.001) (0.001)

District RIGHT MARGIN -0.005 0.006 0.014(0.009) (0.017) (0.018)

District SECOND -0.941 0.538 0.794(0.614) (1.003) (1.059)

Economic SKILL PARTY LEVEL -33.058 -15.334 -30.286 -35.350(84.592) (91.136) (84.120) (82.461)

Control CATHOLICS 0.001 -0.003 -0.006(0.007) (0.007) (0.008)

Control NON AGRICULTURAL 0.023* 0.021 0.021(0.013) (0.013) (0.015)

CONSTANT 18.849 7.646 15.857 20.308(50.160) (54.107) (49.904) (48.898)

ln(σ2j ) -0.041 0.094 -0.101 -0.212

(0.860) (0.854) (0.878) (0.942)Regional Fixed Effects � � � ✓CPC 94.72% 94.01% 95.07% 96.13%N 284 284 284 284�� -50.606 -47.832 -45.984 -41.033p(χ2) 0.000 0.004 0.003 0.006BIC 129.456 146.505 154.107 172.450

* p < 0.10, ** p < 0.05, *** p < 0.01, McFadden’s R2, E(y) = 0.553

Appendix A6: Model Estimation on Non-Left Politicians

The final robustness check presented in this appendix is based on a sub-sample. We re-estimatethe models from Table 2 on a restricted sample that includes only non-left politicians in the Re-

38

ichstag. The first Rokkanian hypothesis formulates hypotheses about the incentives of politicianson the right that face a Social Democratic threat on decision of the latter to support proportionalrepresentation. These empirical tests evaluate the relative effects of seat vote disproportionalityand social democratic threat on the political decisions of right wing politicians only. Models 1eto 4e show the results if this analysis. The results demonstrate that both Rokkanian hypothesesjointly explain the variation in the support for proportional representation among politicians onthe political right. The choice of electoral system can be explained by a combination of seat votedisproportionality and left threat.

Table 11: Replication of Table 2 with Non-Left Members of the Reichstag

Model 1e Model 2e Model 3e Model 4e

SEAT VOTE DISPR. -0.339*** -0.610*** -0.617*** -0.636***(0.044) (0.100) (0.103) (0.118)

SOCIALVOTE 0.017 0.024 0.080* 0.090*(0.011) (0.015) (0.046) (0.053)

SECOND*SOC.VOTE -0.074* -0.094**(0.040) (0.048)

RIGHT MARGIN*SOC.VOTE -0.001 -0.001(0.001) (0.001)

RIGHT MARGIN -0.005 0.005 0.015(0.009) (0.016) (0.017)

SECOND -1.469** -0.100 0.223(0.630) (0.922) (0.970)

SKILL -4.300 -3.980 -3.643 -3.927(3.336) (4.032) (4.009) (4.722)

CATHOLICS 0.017** 0.015** 0.013(0.007) (0.007) (0.008)

NON AGRICULTURAL 0.027** 0.024* 0.024(0.013) (0.014) (0.016)

CONSTANT 2.391 1.203 0.394 2.178(2.020) (2.615) (2.702) (3.163)

Regional Fixed Effects � � � ✓CPC 93.26% 93.78% 92.23% 95.85%N 193 193 193 193�� -56.105 -46.240 -44.172 -37.719p(χ2) 0.000 0.000 0.000 0.000BIC 133.261 134.582 140.971 154.377

* p < 0.10, ** p < 0.05, *** p < 0.01, McFadden’s R2, E(y) = 0.653

39

Appendix A7: Party Defection

In section 4.2 we model the conditions under which politicians defect from the party line andsupport the introduction of proportional representation, even though their respective party opposesthis change in electoral system. In Table 12 we re-estimate the models for a sub-sample that includesonly members of Zentrum.

Table 12: Logit Models on Party Defection for Zentrum

Model 9 Model 10 Model 11 Model 12 Model 13SOCIALVOTE 0.093** 0.118** 0.110**

(0.040) (0.053) (0.052)SKILL LEVEL 0.019 -0.145 -0.087

(0.045) (0.204) (0.228)RIGHT MARGIN 0.007 -0.018 0.010

(0.027) (0.026) (0.028)NON AGRICULTURAL 0.000 0.107 0.047

(0.031) (0.107) (0.120)CATHOLICS -0.061 -0.026 -0.064

(0.043) (0.043) (0.044)SECOND -2.857 -2.208 -2.740

(1.932) (1.779) (1.939)CONSTANT -3.768*** -2.929* 0.888 -0.093 1.417

(0.885) (1.689) (3.493) (3.834) (3.945)N 72 72 72 72 72�� -14.039 -16.593 -1.913 -5.542 -0.307p(χ2) 0.020 0.677 0.210 0.589 0.320BIC 36.632 41.739 29.487 36.744 30.551Standard errors in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01 – Logit modelwith penalized maximum likelihood function, “firthlogit” (Firth, 1993).

To further probe the robustness of these results, we also estimate models using an alternativemeasure of skill level (average skill of total labor force). The results are unchanged with respect tostatistical significance and magnitude of the effects Table 13. Again, the only significant explana-tory factor affecting defection from party line among politicians of the Zentrum is the strength ofleft candidates in a district. Finally, in Table 14 we present the models again but rely on linearprobability model instead of a probit or Firth’s logic model. As with all other model specifications,here again, the strength of left contenders in a given district explains why members of Zentrumdefect from party line and support the introduction of proportional representation.

40

Table 13: Logit Models on Party Defection with Alternative Skill Measure

Model 9 Model 10 Model 11 Model 12 Model 13SOCIALVOTE 0.092** 0.118** 0.110**

(0.040) (0.052) (0.052)SKILL LEVEL 0.019 -0.145 -0.072

(0.045) (0.203) (0.230)RIGHT MARGIN 0.006 -0.019 0.008

(0.026) (0.026) (0.027)NON AGRICULTURAL 0.002 0.108 0.043

(0.032) (0.108) (0.123)CATHOLICS -0.058 -0.019 -0.058

(0.043) (0.042) (0.044)SECOND -3.116 -2.127 -2.905

(1.976) (1.763) (1.980)CONSERVATIVE -2.911* -1.911 -5.336 -2.445 -5.280

(1.599) (1.485) (3.549) (3.147) (3.675)Constant -3.754*** -2.934* 0.554 -0.710 0.812

(0.883) (1.689) (3.522) (3.696) (3.907)N 108 108 108 108 108�� -14.871 -17.416 -2.401 -6.009 -0.709p(χ2) 0.040 0.380 0.202 0.556 0.309BIC 43.789 48.878 37.577 44.793 38.874Standard errors in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01 – Logit modelwith penalized maximum likelihood function, “firthlogit” (Firth, 1993).

Table 14: Linear Probability Model on Party Defection

Model 9 Model 10 Model 11 Model 12 Model 13SOCIALVOTE 0.005** 0.006** 0.005**

(0.002) (0.002) (0.002)SKILL -1.038 -1.048 -0.826

(0.658) (0.671) (0.668)RIGHT MARGIN -0.001 -0.002 -0.001

(0.001) (0.001) (0.001)NON AGRICULTURAL 0.001 0.002 0.001

(0.002) (0.002) (0.002)CATHOLICS -0.000 -0.001 -0.001

(0.001) (0.001) (0.001)SECOND -0.163** -0.129* -0.167**

(0.071) (0.070) (0.071)CONSERVATIVE -0.112** -0.076 -0.116 -0.111 -0.111

(0.048) (0.046) (0.106) (0.108) (0.106)Constant 0.026 0.705* 0.051 0.763* 0.566

(0.038) (0.395) (0.162) (0.445) (0.447)R2 0.070 0.052 0.116 0.090 0.130N 108 108 108 108 108�� 9.840 8.796 12.596 10.987 13.416p(χ2) 0.022 0.061 0.048 0.140 0.047BIC -5.633 -3.546 7.584 10.801 10.625Standard errors in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01 – Logit modelwith penalized maximum likelihood function, “firthlogit” (Firth, 1993).

41

Tab

leA

8:Sk

illed

and

Uns

kille

dw

orke

rsin

1895

–“D

iege

lern

ten

und

unge

lern

ten

Arb

eite

rim

Jahr

e18

95”

Sec

tor

Typ

eC

ateg

ory

Des

cr.8

Ski

llM

9Ski

llF10

Ski

lled

Unsk

illM

11

Unsk

illF12

Unsk

ille

d13

III.

Min

ing,

Sal

twor

ks

IIIa

)O

rem

inin

gO

rem

inin

gB

142

876

342

879

2431

335

2327

836

and

smel

ting

oper

atio

ns

IIIb

)Ir

onst

eelw

orks

Iron

Ste

elW

orks

B2

2291

418

2293

211

2400

4210

1166

10

IIIb

)Sal

tm

ines

Sal

tM

ine

B3

2853

528

5863

4998

6447

IIIc

)Exp

loitat

ion

ofro

cks,

Roc

ksan

dbro

wnco

alB

418

4186

718

4193

1243

2660

7613

0402

bro

wnco

al,oi

l,am

ber

IIIc

)Exp

loitat

ion

ofpea

tExp

loitat

ion

ofpea

tB

517

91

180

4915

1328

6243

IVSto

nec

utt

ing

IVa)

Sto

nem

ason

rySto

nem

ason

ryB

648

373

4048

413

3491

254

3745

Quar

ryw

orke

rsB

719

526

200

1972

643

488

1766

4525

4

Mak

erof

del

icat

est

one

B8

4102

3041

3219

0451

024

14

obje

cts

IVb)

Wor

kers

with

grav

el,sa

nd

Wor

kers

with

grav

el,sa

nd

B9

3746

7438

2029

343

1644

3098

7

Mak

erof

cem

ent

obje

cts

B10

902

1191

327

3415

828

92

IVc)

Wor

kers

with

loam

and

clay

Wor

ker

with

loam

and

clay

B11

232

323

532

4914

933

98

IVd)

Pot

ters

,B

rick

sB

rick

mak

erB

1210

360

5410

414

1499

0212

653

1625

55

Pot

ter

mak

erB

1314

721

122

1484

364

5012

2976

79

Mak

erof

refined

pot

ter

B14

669

4474

311

4326

614

09

Por

cela

info

rmer

B15

1919

340

8223

275

1075

271

1417

866

IVe)

Gla

spro

cess

ing

Gla

sw

orks

B16

1727

418

217

456

1206

233

0615

368

Gla

sfinis

hin

gB

1767

8058

073

6071

293

716

49

Mirro

rgl

asan

dm

irro

rB

1825

4566

632

1118

9459

024

84

fabrica

tion

Toy

sm

ade

out

ofst

one,

B19

694

266

960

164

125

289

por

cela

inor

glas

VM

etal

pro

cess

ing/

Va)

Nob

lem

etal

sG

oldsm

iths,

Jew

eler

sB

2015

418

4368

1978

672

448

712

11

Met

allu

rgy

Oth

erpre

ciou

sB

2150

8922

7773

6689

325

0834

01

met

alpro

cess

ing

Vb)

Non

-nob

lem

etal

sC

opper

smiths

B22

9167

1791

8449

986

585

with

the

exce

ption

Bra

ssfo

under

B23

5569

455

7367

339

712

ofIr

onan

dSte

elT

info

under

B24

1301

2813

2940

922

463

3

Pro

duce

rsof

met

alto

ysB

2519

610

730

323

457

981

3

Oth

erpro

duce

rsof

non

-B

2611

112

489

1160

139

6819

5559

23

pre

ciou

sm

etal

s(e

x.iron

)

Vc)

Iron

and

Ste

elB

rass

cast

erB

2732

432

419

419

213

Wor

ker

inbro

nze

B28

7472

8275

5414

8098

524

65

Met

alfo

under

B29

6588

263

6851

5566

2117

7683

Iron

founder

B30

4181

457

4187

126

396

2000

2839

6

Con

tinu

edon

Nex

tPag

e...

42

Bla

ckan

dB

3155

33

556

2097

200

2297

Plu

mber

B32

4464

346

4468

910

3630

613

42

Tin

wor

ker

B33

3833

215

4048

5468

3912

9380

Mak

erof

nai

lsB

3414

881

1489

288

5234

0

Sec

tor

Typ

eC

ateg

ory

Des

cr.

Ski

llM

Ski

llF

Ski

lled

Unsk

illM

Unsk

illF

Unsk

ille

d

Mak

erof

wires

B35

2320

323

2323

7314

325

16

Mak

ers

ofsc

rew

sB

3633

2952

3381

6348

1575

7923

and

chai

ns

Bla

cksm

ith

B37

1271

5253

1272

0534

0216

135

63

Loc

ksm

ith

B38

2592

9416

425

9458

8318

626

8944

Knife

smith

B39

1195

521

1197

647

5451

252

66

scis

sor

shar

pen

erB

4051

3613

5149

613

7368

6

File

cutt

erB

4149

1417

4931

740

6380

3

Mak

erof

smal

lB

4242

6812

943

9752

4874

559

93

iron

pie

ces

Pro

duce

rof

nee

dle

s,B

4348

0947

052

7930

6024

2454

84

wire

obje

cts

Pro

duce

rsof

writing

B44

547

6113

541

855

3

tool

sm

ade

ofst

eel

VI

Mac

hin

eto

ols

VI

a)Pro

duce

rsof

mac

hin

esPro

duce

rof

mac

hin

eB

4563

089

217

6330

649

611

2792

5240

3

tool

s

VI

b)

Builder

ofm

ills

Builder

ofm

ills

B46

2342

223

4428

213

295

VI

c)B

uilder

sof

Whee

lwrigh

tB

4744

875

2344

898

1100

316

1416

carr

iage

san

dsh

ips

Car

t/ca

rria

gebuilder

B48

3530

1135

4144

8211

846

00

VI

d)

Gunsm

ith

Ship

builder

B49

1001

410

014

1008

022

1010

2

Gunsm

ith

B50

2846

2846

755

80

Wor

ker

ingu

nfo

undries

B51

2133

321

3641

8027

044

50

VI

e)W

atch

mak

ers

Wat

chm

aker

B52

1415

480

114

955

1560

759

2319

VI

f)M

usi

calin

stru

men

tsi)

Pro

duce

rof

artifici

alB

5338

0420

3824

1386

255

1641

(exc

ludin

gm

usi

calto

ysfe

rtiliz

ers

for

childre

n)

ii)

Man

ufa

cture

rsof

B54

5213

371

5584

1698

681

2379

other

musi

calin

stru

men

ts

VIg

)M

athem

atic

alan

diii)

Pro

duce

rof

B55

1864

923

918

888

916

616

1532

surg

ical

inst

rum

ents

surg

ical

tool

s

Vih

)Lam

ps

Lam

pm

aker

B56

435

644

124

2413

1637

40

Vii)

Ele

ctrica

lm

achin

esM

aker

ofB

5734

8012

3492

5669

1066

6735

elec

tric

alm

achin

es

VII

Chem

ical

Indust

ryV

IIa)

Chem

ist

Chem

ist

B58

2912

109

3021

2420

622

1626

422

Con

tinu

edon

Nex

tPag

e...

43

VII

b)

Dru

ggis

tD

rugg

ist

B59

5921

759

2821

4437

625

20

VII

c)Pro

duce

rof

Pro

duce

rof

B60

3422

254

3676

1003

224

3112

463

colo

red

mat

eria

lsco

lore

dm

ater

ials

VII

d)

Pro

duce

rof

Pro

duce

rof

B61

1339

241

1580

8577

7727

1630

4

explo

sive

mat

eria

lsex

plo

sive

mat

eria

ls

VII

e)Pro

duce

rof

i)Pro

duce

rof

B62

354

335

782

8362

08

artifici

alfe

rtiliz

ers

artifici

alfe

rtiliz

ers

ii)

Knac

ker

B63

373

137

422

250

272

Sec

tor

Typ

eC

ateg

ory

Des

cr.

Ski

llM

Ski

llF

Ski

lled

Unsk

illM

Unsk

illF

Unsk

ille

d

VII

ISec

ondar

ypro

-V

IIIa

)Sec

.prd

cof

frst

rC

har

coal

burn

erB

6426

526

554

519

564

duct

sof

fore

stry

(Soa

ps,

Gas

B65

769

477

311

238

102

1134

0

Fat

s,O

ils,

Var

nis

hes

)V

IIIb

)Soa

p-m

akin

gSoa

pm

aker

sB

6617

6311

718

8041

9119

0961

00

VII

Ic)

Oil

mills

Oil

mills

B67

919

1693

537

1725

139

68

VII

Id)

Oils,

varn

ishes

Wor

kers

with

oils

,B

6894

613

710

8345

1212

3757

49

VII

Id)

Oils,

varn

ishes

varn

ishes

IXTex

tile

indust

ryIX

)Pre

par

atio

nof

Pre

par

atio

nof

B69

1185

811

1996

6679

6913

1359

2

spin

nin

gm

ater

ials

spin

nin

gm

ater

ials

IXa)

Spin

ner

Spin

ner

B70

2357

033

397

5696

738

085

6206

810

0153

Clo

thm

aker

B71

1994

310

552

3049

514

481

1381

528

296

IXb)

Wea

ver

Wea

ver

B72

1026

3387

322

1899

5549

661

5574

110

5402

IXc)

Rib

bon

man

ufa

cture

rR

ibbon

man

ufa

cture

rB

7342

838

681

428

624

553

1

IXd)

Knitte

rK

nitte

rB

7417

053

1996

237

015

3925

5983

9908

IXe)

Em

bro

ider

erEm

bro

ider

erB

7537

7169

4010

711

1426

5727

7153

Ixf)

Dye

rD

yer

B76

1608

954

316

632

1367

232

2816

900

Ble

acher

B77

1033

580

9418

429

1267

310

505

2317

8

Ixg)

Fringe

mak

erFringe

mak

erB

7857

5751

8010

937

3392

4580

7972

Ixh)

Rop

em

aker

Rop

em

aker

B79

4915

195

5110

1881

1710

3591

Sai

lmak

erB

8072

625

397

920

460

380

7

XPap

erIn

dust

ryX

)Pap

eran

dPap

eran

dca

rd-

B81

8031

2540

1057

134

650

1888

753

537

card

boa

rdm

aker

sboa

rdm

aker

s

Toy

sm

anufa

cture

red

B82

1085

440

1525

130

192

322

from

pap

erm

ache

Xb)

Boo

kbin

der

Boo

kbin

der

B83

2695

425

0529

459

4563

1225

816

821

XI

Lea

ther

wor

kin

gX

Ia)

Wor

kers

with

leat

her

Lea

ther

Indust

ryB

8428

72

289

178

2320

1

Tan

ner

B85

2411

187

924

990

1206

689

012

956

Lea

ther

colo

rer

B86

1239

4112

8025

7620

427

80

XIb

)O

ilcl

oth,ru

bber

Oilcl

oth

B87

425

2745

213

9914

715

46

Con

tinu

edon

Nex

tPag

e...

44

Pro

duce

rof

B88

477

147

624

6025

3116

9141

rubber

obje

cts

Pro

duce

rsof

toys

man

u-

B89

41

518

085

265

fact

ure

red

out

ofru

bber

Man

ufa

cture

rof

toys

B91

206

230

436

5818

224

0

mad

eou

tof

leat

her

XIc

)Sad

dle

mak

ers

Sad

dle

Mak

erB

9037

903

589

3849

215

3812

0827

46

Wal

lpap

erer

B92

1940

124

719

648

634

276

910

XII

Woodw

orkin

gX

IIa)

Woo

dw

orki

ng

Woo

dw

orke

rsB

9315

461

8315

544

2934

183

130

172

Man

ufa

cture

rof

B94

7458

277

7735

5339

1442

6781

larg

ew

ood

obje

cts

Sec

tor

Typ

eC

ateg

ory

Des

cr.

Ski

llM

Ski

llF

Ski

lled

Unsk

illM

Unsk

illF

Unsk

ille

d

XII

b)

Man

ufa

cture

rsof

Cab

inet

mak

erB

9523

5594

240

2358

3467

0482

275

26

pol

ished

woo

dob

ject

s

XII

b)

Bar

relm

aker

sB

arre

lm

aker

B96

2999

723

3002

013

8568

1453

XII

d)

Bas

ket

mak

erB

aske

tm

aker

B97

1152

915

2213

051

1174

393

1567

XII

e)W

eave

rof

stra

wO

ther

wea

ver

ofB

9814

0621

1235

1888

410

1819

02

woo

dan

dst

raw

XII

f)Turn

erTurn

erB

9919

400

3319

433

595

161

756

Mak

erof

woo

dto

ysB

100

1758

519

2277

804

735

1539

Mak

erof

cork

obje

cts

B10

186

7977

694

5522

5313

2235

75

XII

g)C

ombm

aker

,bru

shm

aker

sC

ombm

aker

B10

297

544

1019

273

129

402

and

mak

ers

ofot

her

woo

dob

ject

s

Bru

shm

aker

B10

368

3211

3379

6519

4622

6142

07

Um

bre

lla

mak

erB

104

1407

1091

2498

1528

666

2194

Mak

erof

pic

ture

and

B10

513

141

497

1363

816

2369

123

14

and

mirro

rfr

ames

XII

IFood

pro

cess

ing

XII

I)Pro

duce

rof

Mille

rB

106

5038

141

5042

215

621

806

1642

7

and

pro

duct

ion

vege

tarian

food

item

s

Bak

erB

107

1144

4847

411

4922

5951

8654

1460

5

Pas

tryb

aker

B10

814

230

351

1458

118

4939

8858

37

Suga

rpro

duce

rB

109

1543

5215

9522

785

3766

2655

1

Pro

duce

rof

other

B11

012

4592

321

6871

9183

2915

520

vege

table

food

item

s

XII

Ib)

Pro

duce

rof

Butc

her

B11

186

303

295

8659

845

4570

1011

555

anim

alfo

oditem

s

Oth

eran

imal

food

item

sB

112

5817

824

6641

3589

2154

5743

Man

ufa

cture

rof

B11

356

321

584

5868

405

6273

min

eral

pro

duct

s

Con

tinu

edon

Nex

tPag

e...

45

XII

Ic)

Drinks

Mal

ter

B11

415

707

1577

1280

5913

39

Bre

wer

B11

535

784

3635

820

3130

814

3832

746

Man

ufa

cture

rof

win

eB

116

4347

1643

6372

6581

080

75

Win

emak

erB

117

3733

2637

5991

866

984

Vin

egar

dis

tiller

B11

825

05

255

851

5590

6

XII

Id)

Tob

acco

Tob

acco

mak

erB

119

4375

641

344

8510

012

323

2334

435

667

XIV

Fas

hio

nX

IVa)

Clo

thes

Sea

mst

ress

B12

097

190

9719

095

795

7

&B

ekle

idung

Tai

lor

B12

113

1486

5884

619

0334

550

415

965

Mak

erof

linen

B12

231

7024

667

2783

722

6430

3552

99

milliner

yB

123

112

1542

915

541

174

1071

1245

Dol

lmak

erB

124

568

1153

1721

137

565

702

Mak

erof

artifici

alflow

ers

B12

563

159

6665

9754

121

3426

75

Hat

mak

erB

126

5712

2690

8402

4087

3032

7119

Bon

net

mak

erB

127

751

276

1027

5668

124

Furr

ier

B12

853

7686

462

4081

351

713

30

Sec

tor

Typ

eC

ateg

ory

Des

cr.

Ski

llM

Ski

llF

Ski

lled

Unsk

illM

Unsk

illF

Unsk

ille

d

Glo

vem

aker

B12

946

7833

7380

5138

194

913

30

Mak

erof

ties

B13

018

717

5819

4512

920

133

0

and

susp

ender

s

Cor

setm

aker

B13

144

643

3447

8026

813

0115

69

XIV

b)

Shoe

sShoe

mak

erB

132

1476

5574

3115

5086

4462

3383

7845

XIV

c)B

arber

Bar

ber

B13

321

920

1821

938

5014

64

Hai

dre

sser

and

B13

482

1449

687

1043

3477

mak

ers

ofw

igs

XIV

d)

Wor

kers

inpublic

Wor

ker

inpublic

bat

hs

B13

531

511

242

717

2713

6330

90

bat

hs

and

clea

ner

sW

asher

B13

655

519

947

2050

221

7526

710

2888

5

Clo

thes

clea

ner

B13

781

1495

5449

103

XV

Con

stru

ctio

nX

va)

Con

stru

ctio

nw

orke

rC

onst

ruct

ion

wor

ker

B13

897

7923

9802

3120

6281

1632

0178

XV

b)

Fie

ldsu

rvey

orFie

ldsu

rvey

orB

139

1467

314

7043

7982

4461

XVc)

Brick

laye

rB

rick

laye

rB

140

3722

3418

237

2416

4454

721

7246

719

XVd)

Car

pen

ter

Car

pen

ter

B14

115

5391

8415

5475

4594

271

4865

Xve

)G

lasi

erG

lasi

erB

142

1076

315

1077

821

936

255

Xvf

)Pai

nter

Pai

nter

B14

392

589

104

9269

324

9423

227

26

XVg)

Stu

cco

wor

ker

Stu

cco

wor

ker

B14

411

276

1111

287

757

4580

2

XV

h)

Roo

fer

Roo

fer

B14

519

512

1951

223

0131

2332

Xvi

)Sto

nem

ason

rySto

nem

ason

ryB

146

1165

313

1166

653

5037

5387

XV

j)M

aker

ofw

ells

Mak

erof

wel

lsB

147

1434

214

3610

6513

1078

XV

k)Plu

mber

Plu

mber

B14

834

774

3481

3224

2432

48

XV

l)G

asst

ove

mak

erG

asst

ove

mak

erB

149

1139

33

1139

677

348

821

Con

tinu

edon

Nex

tPag

e...

46

XVm

)C

him

ney

swee

per

Chim

ney

swee

per

B15

053

5153

5121

324

XV

IPri

nti

ng

and

XV

Ia)

Woo

dca

rver

Woo

dca

rver

B15

131

8143

3224

283

312

595

phot

ogra

phy

XV

Ib)

Boo

kprint

erB

ookp

rint

erB

152

5188

414

6153

348

5610

7191

1280

1

Lithog

rapher

B15

316

646

353

1699

914

7225

9340

65

Lithog

rapher

usi

ng

B15

487

922

901

9826

336

1

copper

and

iron

B15

487

922

901

9826

336

1

Col

orprint

erB

155

489

294

783

182

490

672

XV

Ic)

Phot

ogra

pher

Phot

ogra

pher

B15

659

4359

965

4224

821

946

7

XV

IIA

rtis

tic

XV

IIa)

Pai

nter

and

carv

erPai

nter

and

carv

erB

157

2130

9222

2215

824

182

occ

upat

ions

XV

IIb)

Engr

aver

Engr

aver

B15

888

8117

090

5139

672

468

XV

IIc)

Cal

igra

pher

Cal

igra

pher

B15

932

2917

334

0224

3054

XV

IId)

Oth

erar

tist

icoc

cupat

ions

Oth

erar

tist

icoc

cupat

ions

B16

014

7219

316

6513

511

825

3

Com

mer

cew

ith

B16

192

812

710

5520

412

6331

2674

3

mer

chan

dis

e

XV

III

Tra

de

XV

III)

Com

mer

cew

ith

good

sShop

pin

gag

ent

C1

1796

11an

dco

mm

erce

8086

626

0477

9697

413

901

1118

75

XV

IIb)

Com

mer

cew

ith

cred

itM

oney

C2

849

4189

040

7875

4153

Cre

dit

C3

773

978

285

4896

8644

Sec

tor

Typ

eC

ateg

ory

Des

cr.

Ski

llM

Ski

llF

Ski

lled

Unsk

illM

Unsk

illF

Unsk

ille

d

XV

IIIc

)C

omm

erce

with

boo

ks,

Com

mer

cew

ith

boo

ks,

C4

5029

473

5502

3518

209

3727

art

incl

udin

gan

tiqu

esar

tin

cludin

gan

tiqu

es

New

spap

erprint

erC

513

542

177

1571

3167

4738

Ped

dle

rC

661

935

897

727

814

842

6

Busi

nes

sm

edia

tor

C7

363

1037

374

821

769

Tra

din

gap

pre

ntic

eC

822

610

236

2696

320

1828

981

Auct

ioner

C9

198

3823

617

6591

1856

XIX

Insu

rance

IXIa

)A

ctuar

yA

ctuar

yC

1028

33

286

1145

3211

77

XX

Tra

nsp

orta

tion

XX

a)Tra

nsp

orta

tion

onla

nd

Drive

rof

per

s.ve

hic

les

C13

700

270

225

703

8625

789

Str

eetc

arw

orke

rC

1455

755

713

843

117

1396

0

Rai

lway

wor

kers

C15

1080

1410

9439

160

237

3939

7

XX

b)

Wat

ertr

ansp

orta

tion

Nav

igat

orC

1617

917

972

412

736

Nav

igat

oron

ship

sC

1770

4570

4550

9880

5178

onse

asan

dco

asts

Nav

igat

oron

rive

rsC

1818

488

2118

509

1384

610

313

949

Lon

gshor

eman

C19

639

639

2420

1224

32

XX

c)Par

celca

rrie

rPar

celca

rrie

rC

2019

214

206

7144

2062

9206

XX

d)

Mor

tici

anM

ortici

anC

2178

785

2115

740

2855

Con

tinu

edon

Nex

tPag

e...

47

XX

IH

otel

san

dre

stau

rants

XX

Ia)

Hot

els

and

rest

aura

nts

Hot

els

and

rest

aura

nts

C22

5237

037

121

8949

142

737

1259

2416

8661

1D

escr

ipti

onan

dcl

assi

ficat

ion

ofth

eIm

peri

alSt

atis

tica

lOffi

ce.

Thi

sis

the

fines

tca

tego

riza

tion

leve

l.2N

umbe

rof

skill

edm

enin

ace

rtai

nca

tego

ry.

3N

umbe

rof

skill

edw

omen

ina

cert

ain

cate

gory

.4N

umbe

rof

skill

edem

ploy

ees

ina

cert

ain

cate

gory

.5N

umbe

rof

unsk

illed

men

ina

cert

ain

cate

gory

.6N

umbe

rof

unsk

illed

wom

enin

ace

rtai

nca

tego

ry.

7So

urce

:“D

iege

lern

ten

und

unge

lern

ten

Arb

eite

rim

Jahr

e18

95”.

1899

.K

aise

rlic

hes

Stat

istis

ches

Am

t.D

iebe

rufli

che

und

sozi

ale

Glie

deru

ngde

sD

euts

chen

Vol

kes

nach

der

Ber

ufsz

ählu

ngvo

m14

Juni

1895

,pag

es77

-80.

48