The Rise of Swedish Social Democracy

Abstract We examine the rise and mobilizational dynamics of social democracy, employing data reported by the Swedish authorities on the distribution of voting eligibility, turnout and partisan vote in local elections during the 1910s at a high level of disaggregation (by narrow income segments and administrative units). In line with the existing literature, we show that electoral socialism depended on both the extension of the suffrage to its ‘natural’ electorate, that is, the urban working class, and the organizational capacity of trade unions and other civic associations. In addition, we show that socialist support was not uniform within the working class – even for a highly homogeneous society. Instead, the social-democratic vote was initially stronger among low- to middle-income workers, only expanding to poor voters later in time. We complement our local data with an interwar panel analysis of socialist vote and post-war survey data.

Although the electoral rise and consolidation of social democracy has been the object of several path-breaking studies, their empirical validation has been marred by the fact that researchers have had to rely on aggregate electoral and social data for the period that preceded the systematic use of surveys.Here, we start to remedy this problem by relying on unique, highly granular data on Swedish local elections conducted between 1910 and 1918.Collected by the electoral administration of Sweden, the data include the number of registered and eligible voters, whether they voted or not, and the party they voted for at a high level of disaggregation: by both local district and income segmentsmost of them defined by Kr. 100 intervals.Matching that (quasi-)individual information with data on trade union density and on membership in free churches and the temperance movement, we determine the economic and organizational covariates of turnout and partisan vote since the introduction of universal suffrage in 1909 and until the abolition of the plural vote in local elections in 1919a decade of fundamental electoral growth for Swedish social democrats.We complement this analysis with interwar aggregate data at the local level on industrial structure, unions and social-democratic vote, and with a survey on party vote conducted for the election of 1960.
Our empirical analysis leads us to make three claimswith the first two integrating the two main theoretical approaches advanced so far to explain the growth of electoral socialism.In the first place, we show that its success depended on the political inclusion of specific social strata, mainly, the industrial working class, which socialist parties defined as their natural constituency (Lipset and Rokkan 1967;Przeworski and Sprague 1986).In the second place, we find that universal suffrage and the concomitant enfranchisement of working-class voters were necessary but not sufficient conditions to propel social democracy to victory.To win, socialist candidates needed strong organizational structures, that is, a mass party and friendly, robust unions, to take previously non-mobilized electors to the polls while moving old voters away from other parties (Bartolini 2000;Katznelson and Zolberg 1986;Kunkel and Pontusson 1998;Luebbert 1991;Rennwald and Pontusson 2019).To these two conditionswhich follow Bartolini's (2000) work at the macro or national level, transposed to and examined at the micro or individual one in this letterwe add a third claim.The process of socialist mobilization followed very specific dynamics, which, though well understood by socialist strategists at that time, have been, to our knowledge, hardly discussed in the academic literature.Social democracy was initially supported by a relatively narrow segment of the electoratemostly relatively affluent industrial workers.It was later on that, building on that initial core, it attracted the vote of poorer, less organized and often previously never mobilized individuals.
The letter is organized as follows.The first section develops our theoretical expectations.The second section discusses the advantages of examining the Swedish case and describes our main electoral data for the 1910s.The third section proceeds to estimate the covariates of participation and party vote.The fourth section looks at the final expansion and consolidation of the socialdemocratic vote using aggregate data until the 1940s and a national survey conducted in 1960.Finally, the fifth section recaps our results, placing them within the more general debate about the rise of electoral socialism.

The Mobilization of Socialist Voters
In nineteenth-century liberal political regimes, which generally restricted the right to vote to male individuals who had sufficient economic means and/or met certain literacy requirements, socialist parties entered politics as anti-system organizations at first.In 1872, the First International proclaimed that 'the organization of the proletariat into a political party is necessary to insure the victory of social revolution and its ultimate goalthe abolition of classes'. 1As such, socialist parties could only hope to capture a narrow segment of voters.Indeed, countries with restrictive franchise conditions elected few, if any, socialist candidates.Socialist parties only grew in parliamentary size after the introduction of male universal suffrage in Belgium in 1893, Norway in 1900, Sweden in 1911, Britain and the Netherlands in 1918, and Germany after the abrogation of anti-socialist repressive legislation in the late 1880s.
The introduction of universal suffrage was a necessary but not sufficient condition to develop large social-democratic parties.In addition to having the right to vote, workers had to both identify social-democratic parties as those best suited to defend their interests and mobilize at the ballot box. 2 The electoral mobilization of low-income voters was a function, in the first place, of their individual resources (Blais and Achen 2010;Rosenstone and Hansen 1993).Within the 'natural' electorate of social-democratic parties, that is, lower-class voters, active support was more likely to happen, at least initially, among relatively affluent workers, who arguably had more personal and interpersonal resources, as well as more time to engage in civic life and in politics, and who worked under conditions (stable jobs in modern factory plants) that probably facilitated collective action and unionization.Voting for socialist candidates depended, in the second place, on seeing them as electorally viable, that is, as having a reasonable chance of entering parliament.The use of majoritarian electoral rulesin place until the first decades of the twentieth centuryreinforced strategic voting and therefore deterred working-class voters from switching to (then) untested socialist candidates.Hence, the development of strong socialist party machines and pro-socialist trade unions was instrumental in bringing low-income voters to the ballot box, either directly (by relying on propaganda, targeted benefits and even coercion) or indirectly (through the informal social networks in which both those organizations and voters were embedded) (Amat et al. 2020;Bartolini 2000;Flavin and Radcliff 2011;Kerrissey and Schofer 2013).
Given the conditions that affected the likelihood of voting, the timing and nature of the mobilization of potential socialist voters followed a very specific pattern.Right around the introduction of universal suffrage, social democracy was supported by a relatively narrow segment of the electorate, which coincided with the relatively better-off industrial worker.Writing in 1898, Eduard Bernstein, a German 'revisionist' socialist politician, stressed that 'in Germany and the other continental civilized lands … the best-paid workmen stand at the head of the class war' (Bernstein 1993: 104).By contrast, support for social democracy among low-income workers (as opposed to low-to middle-wage earners) was much lower, developing only over time. 3It was therefore at the end of this mobilizational process that social democracy adopted the nature of a fully workers' party that the literature has so often emphasized as characterizing electoral socialism.
The Case of Swedish Social Democracy Early twentieth-century Sweden provides us with a valuable setting to examine the conditions that led to the rise of electoral socialism due to its political and institutional context.First, the electoral reform of 1909 expanded the franchise sharply from about 30 per cent of all men over 24 years old in 1905 to almost 80 per cent, making it next to impossible for non-socialist parties to shift leftwards to compete for the new electorate in a credible way and enabling us to assess the reaction of voters (both old and new) to what were stable policy programmes. 4econdly, the concomitant introduction of proportional representation attenuated the incentive to vote strategically.Accordingly, the actual support for the Swedish Social Democratic Workers' Party (Sveriges socialdemokratiska arbetareparti [SAP]) can be interpreted as revealing something close to the true preferences of voters (or at least closer to them relative to voting under majoritarian rules).Thirdly, even though there was a medium-sized evangelical movement, Sweden's relatively high religious and cultural homogeneity minimizes the noise that other factors may have played in explaining the success of socialism.Finally, our data, to be described shortly, cover the period during which the SAP experienced most of its growth.After the electoral reform of 1909, the socialist vote more than doubled in size to 29 per cent of the vote in 1911.Three years later, the SAP's vote share increased to 36 per cent.Afterwards, it grew at a slower paceto 40 per cent of the votes in 1924 and over half of the ballots in 1940 (Tingsten 1973).
In addition to national elections, Sweden held elections to choose the governing bodies of primary communes and secondary communes (landsting kommuner), according to the Municipal Administration Act 1862.Except for the five largest cities of Stockholm, Göteborg, Malmö, Norrköping and Gäfle, the primary communes were embedded in secondary communes.The latter, which generally corresponded to the Swedish län ('provinces'), were governed by separate general councils (Sandalow 1971).
After the reform of 1909 and until 1918, general councils were elected through direct elections, proportional representation and plural voting.Voters received as many ballots as their taxable income according to a graduated scale ranging from one to forty votes.Electoral districts were classified as either rural or urban.In rural districts, voters were given one vote for each additional 100-crowns taxable income bracket up until 1,000 Swedish crowns and then another ballot for each additional 500-crowns income segment up to a maximum of 40 ballots.In urban districts, voters received one ballot for each 100-crowns segment up until 2,000 crowns and then one vote 3 For a reference to the low mobilization of poor voters in the literature on democratization, see Ansell and Samuels (2014, ch.2).
https://doi.org/10.1017/S0007123422000102Published online by Cambridge University Press for each additional 500-crowns income segment, again up to a maximum of 40 ballots.As described in Online Appendix A, taxable income differed slightly from all income but essentially tracked it.
Electors included men and unmarried women with (taxable) income, as well as some corporations.We only have data on the distribution of registered voters by type (gender and whether it was an individual or a corporation) for the city of Stockholm.There, for 1913-14, 72.9 per cent of the registered electors were men, 23.8 per cent were unmarried women and 3.3 per cent were corporations. 5 Elections to the general councils took place every two years from 1910 to 1918.The 1910 election encompassed all of Sweden.Afterwards, each election was held in half of the electoral districts of each län, resulting in the full renovation of each general council every four years.For each secondary commune and income bracket, the Swedish Central Bureau of Statistics published the number of registered and eligible electors (we later discuss eligibility conditions), the number of people that turned out to vote, and the ballots cast for each party.Those income-segmentcommune units were small.A total of 90 per cent of all units (again, every separate income bracket for each secondary commune holding elections in a given year) had fewer than 300 electors.The average unit had fewer than 150 electors.
In the analysis that follows, we take the very fine-grained data for the elections to general councils to be a good strategy to approximate general electoral behaviour in Sweden at the time of its transition from a semi-open to a fully open polity for two reasons.In the first place, suffrage in the local elections was fairly wide. 6In the electoral cycle of 1912-14, for example, there were 955,510 registered voters in rural districts and 335,534 in urban districts (119,649 of them in the city of Stockholm)equivalent to about 23 and 29 per cent of total population, respectively.In the national election of March 1914, where only men could vote, there were 1,034,161 individuals registered in the countryside and 351,686 in towns.Since there was no income qualification to vote in national elections after 1909, the national electorate was logically larger than the communal electorateespecially when we exclude female voters from the latter.In the second place, the district-level number of eligible voters and party votes were tightly correlated (with r 2 values of about 0.9) between local and national elections (for a full analysis, see Online Appendix B).
Figure 1 plots the proportion of registered voters in the 1912-14 local elections (with the exclusion of the cities of Stockholm, Göteborg, Malmö, Norrköping and Gäfle, for which we have no data) by income bracket and by type of district (rural or urban).A total of 54 per cent of all registered voters had a taxable income equal to or lower than 400 crowns.The top 10 per cent of the population reported a taxable income of 2,500 crowns or more.The top 1 per cent started at 8,500 crowns or morea figure not shown in the graph.Rural incomes were considerably lower than urban incomes.The median taxable income was about Kr. 300 in rural districts and Kr.700 in urban areas.The taxable income in the 90th percentile was about Kr. 1,400 and Kr.3,600, respectively. 7igures 2A and 2B show the proportion of registered electors that were eligible by taxable income segment for each electoral cycle (1910, in urban and rural areas, respectively.The income axis gives the cumulative distribution function of the registered population (all men and unmarried women with some revenue).For example, 50 per cent of registered voters in urban districts had a taxable income of 600 Swedish crowns or less.Eligibility to vote, which was denied to anyone who was a foreigner, had a criminal record, received public assistance, was bankrupt, had not paid taxes or was serving in the military, rose with income.In 1910, 5 Data for the city of Stockholm and the year of 1913 come from the Stockholms stads statistik kontor (1915, Table 78).The Swedish Central Bureau of Statistics did not include any data on local elections for the cities of Stockholm, Göteborg, Malmö, Norrköping and Gäfle in its volumes on landsting kommuner elections (Statistiska centralbyrån 1912(Statistiska centralbyrån , 1914(Statistiska centralbyrån , 1916(Statistiska centralbyrån , 1918)).while only two thirds of all low-income registered voters in urban districts and three quarters in rural areas were eligible, 95 per cent of those at the top 1 per cent were.
Figures 2A and 2B also report turnout rates (over registered voters) by taxable income and electoral cycle.Turnout differentials across income were extremely sharp.In urban areas, less than 25 per cent of the bottom half of the income distribution cast a ballot in 1910three times lower than among the top 10 per cent.In rural districts, participation among low-income voters was higher, but there were still important inequalities along the income distribution.Turnout hardly changed throughout the 1910s: participation experienced a small spike in 1912-14 but then fell again in 1916-18to the levels of 1910.
Figures 3A and 3B plot the proportion of eligible voters voting for the Conservatives, Liberals and Socialists.Support for the SAP had three characteristicsin line with our theoretical expectations.First, it had a hump-shaped structure.It was extremely low among the very poor and tied with or below support for the Liberals and Conservativessomething that may have been related to the presence of clientelistic voting for the latter.Support for the SAP then grew with income, peaking at about 20 per cent among eligible individuals (and therefore over a third of all ballots cast) at the 60th percentile of the income distribution, which, according to Bagge et al (1933), coincided with the earnings of skilled manufacturing workers.8It then fell quickly to less than 5 per cent for the top decile, where conservatives gathered an overwhelming majority of votes.Secondly, the social-democratic vote was urban in nature.Although the SAP vote also exhibited a concave structure in rural districts, overall support was much lower than in cities in 1910about half of the urban vote among lower-middle-and middle-income strata.Last but not least, social democracy grew everywhere over time.By 1916-18, the SAP had overcome both the Liberals and Conservatives among the bottom 75 per cent in cities.In rural areas, where it absorbed part of the Liberal vote and vied for first position with the Conservatives through a great part of the income distribution, support for the SAP caught up with the level of vote in urban districts.Still, it was highly dependent on middle-income voters, which were arguably manufacturing workers.Among those voters earning less than Kr.400, a figure corresponding to the average earnings of male farm workers (Bagge, Svennilson, and Lundberg 1933), the SAP vote remained below 10 per cent.

Empirical Analysis
We proceed now to examine the covariates of abstention and partisan vote throughout the 1910s.To do so, we transform our data into individual observations, assigning to each one the mean income of their taxable income segment. 9Our baseline estimation model is as follows:  : 1910, 1912-14 and 1916-1918) (raw data).

9
As an example of our procedure, consider the district of rural Uppsala in 1910.In income group 5 (Kr.400 to 499), 352 people voted Liberal, 259 people voted Conservative, 112 voted Socialist and 276 abstained.We create a vector of 999 individual observations, with 352 Liberals, 259 Conservatives, 112 Socialists and 276 non-voters.To each of them, we assign the mean income of rural income group 5 (Kr.450).Although the assignment procedure adds some noise to the estimations, it is low given how narrow income segments are.
where: S it represents turnout or party vote of elector i; logY i is the logged taxable income of elector i; UNION i is the ratio of trade union members over workers in the district of i; FREE i is the percentage of the population in Protestant non-Lutheran denominations; TEMP i is the percentage of members of teetotaller associations over population in the district of i; GINI (the Gini coefficient) is the income inequality in the district of i. 10 The model also includes district dummies (D n ), a rural versus urban district dummy (R n ) and year dummies (Z j ).We employ a multinomial logit model to estimate the likelihood of a given outcome under the assumption that voters chose among an array of four excluding alternatives: abstaining or voting for one of three parties (the Conservatives, Liberals or Socialists).

Income and Vote
We expect income to increase personal resources and, therefore, the probability of participating in the election.Income and the probability of voting for the Socialists should then be related through a concave function as a result of two forces.By raising the mobilization of voters, income should have a positive mechanical effect on voting for the Socialists.However, at some income threshold, a higher income should be associated with a higher likelihood of rejecting the redistributive platform of the SAP.We model these two effects through income and income squared.

Trade Unions
Swedish unions, which expanded continuously (with the exception of a temporary decline that followed a failed general strike in 1909), only supported Socialist candidates (Therborn 2010).Accordingly, we expect the size of trade unions in each district to correlate positively with turnout and the Socialist vote.

Teetotalism
In addition to trade unionism, the Swedish polity had two broad social movements at the turn of the century: the temperance movement and free churches.The former, which emerged in the 1870s, gradually changed from emphasizing alcoholism as an individual problem to demanding legislation to restrict the sale of alcoholic beverages.In response to the opposition of beer producers, the Agrarian Association, the Conservative Party and business associations, teetotalism provided Liberal candidates with key organizational resources (Hurd 1994: 48).Over time, the relationship between social democracy and the temperance movement tightened too.A total of 85 per cent of Socialist parliamentarians and more than half of the 1908 SAP congress delegates were teetotallers.In Sweden's rapidly emerging factory towns, temperance organizations offered structured forms of leisure that went beyond drinking and card-playing, supplying everyone with both intellectual 'resources for the self-improving worker' (Hurd 1994: 40) and organizational skills that would become crucial to develop unions (Sandell 2001).

Free Churches
Although the near majority of the Swedish population belonged to the Lutheran state church, several free churches emerged in the mid-nineteenth century.As in other Protestant countries, their members opposed the Conservative Party, which was aligned with the state church, and supported progressive politicians.However, because the 'free church movement accused the 10 The data for union, free churches and temperance membership is for 1910 (for the 1910 election), 1913 (for the elections of 1912-14) and 1917 (for the elections of 1916-18).For the year 1918, we add the votes for the Left Socialist Party to the votes of the SAP and the votes for the National Farmers' Union and to the Farmers' League to the votes of the Conservative Party to make the analysis comparable across elections.temperance movement of being too close to the socialist labor movement's "subversive" ideas' (Sandell 2001), its numbers should be weakly, if at all, related to the Socialist vote.

Estimations
Online Appendix D (Tables 1-6) reports the full estimations.With over 2.5 million observations, coefficients are precisely estimated.Given the logistic structure of the model, and to ease the interpretation of the models, Figure 4 graphs the predicted probability of turning out and of voting for the SAP by level of income (estimated on a logarithmic scale) over time in urban and rural districts separately. 11Probabilities are estimated for eligible voters (not conditioned by turnout). 12 Turnout increased with income.The probability of voting went from around 0.25 for a taxable income of 100 crowns to 0.75 for one of 1,200 crowns.Afterwards, the rise of participation flattened for incomes over 10,000 crownswith a turnout probability of 0.9.In line with our expectations, support for the SAP exhibited a concave structure, rising with taxable income up to around 500 crowns in urban districts and 1,000 crowns in rural districts, and then declining.Over time, the share of the Socialist vote grew, particularly among low-income votersin parallel with rising electoral participation.
Figure 5 displays the probability of abstention and the Socialist vote as a function of different income levels and of trade union density. 13Different lines depict the impact of income at the 10th, 25th, 75th and 90th percentiles of the income distribution.Trade union density varies along the horizontal axis.Abstention was higher among poorer voters.It should also be noticed that for the bottom half of the income distribution, union density was positively correlated with abstention.The result probably points to a compositional effect: more unionized districts had a higher concentration of poorer voters. 14 In line with Figure 4, Figure 5 shows that the relationship between income and vote for the SAP had a convex structure: support was highest for middle percentileswhere lay Sweden's industrial labour aristocracy, according to Bagge, Svennilson, and Lundberg's (1933) data.Union density had a multiplicative effect over income in the latter's effect on voting for the SAP.The probability of casting a vote for the Socialist list was slightly over 10 per cent for a lowincome voter in both unionized and non-unionized districts.By contrast, a voter at around the 75th percentile (approximately, a skilled manufacturing worker) was twice as likely to vote for the SAP in a highly unionized district than in a district without unions.
Online Appendix D plots the effect of both the temperance movement and membership in free churches on the SAP vote.The strength of the teetotaller movement in the elector's district was associated with an average decline of 10 percentage points in abstentionwith a much stronger effect among poor voters.The presence of temperance associations covaried with weaker support for the SAPbut the effect was very small and only affected high-income voters (see Figures D1.A and D1.B in the Online Appendix).Membership in free churches was positively correlated with the SAP vote in models with logged incomequestioning the literature that has emphasized the opposition of evangelical movements to socialism.The effect becomes weaker when we drop 11 The results do not depend on using this functional form (see Online Appendix E). 12 To plot variation over time, Figure 4 is based on the following estimation: It is also estimated separately for urban and rural districts.Note: The predicted probabilities are estimated separately for urban and for rural districts, and are calculated using the following model: published census data containing socio-economic characteristics whose association with electoral behaviour can be explored: broad economic sectors (agriculture, industry and tertiary, sometimes divided into a few subsectors); and occupational categories.Although the latter could be used to construct social classes, that data are only available at the län level (24 observations) and therefore not sufficiently disaggregated to permit plausible statistical tests.Hence, we employ the sectorlevel data, which come at either the county level (over 400 observations, for 1920 and 1930) or the municipal level (around 2,500 units, for 1910 and 1940), matching them to electoral results, to trace the growth of the SAP. Figure 6 reports the ecological inference estimates (following King's [1997] method) of the proportion of both industrial and agricultural employees abstaining and voting for the SAP  and the whole Left. 16Online Appendix F details the sources and estimation methods.Among individuals employed in the industrial sector, there was a clear process of socialist mobilization (see Figure 6A).Abstention halved from 1911 to 1940.Support for the SAP went from 30 per cent of eligible voters in 1911 (where eligibility was not equivalent to adulthood) to almost 50 per cent in 1921 (with full universal suffrage) and almost 70 per cent in 1940.From 1932 to 1940, the SAP absorbed almost all the vote of left-wing parties.Support in the agricultural sector was much lower but it also grew over timeto almost 20 per cent in 1940.

Socialist Growth and Trade Unions
To estimate the impact of unionization on changing support for the SAP, we employ panel data of fifteen elections (from 1917 to 1944), with the following functional form: where: S stands for socialist share of the vote; UNION i is the ratio of trade union members over male eligible voters in district i; L n indicates län dummies; and Z j corresponds to year dummies. 17Results are reported in Table 1.The even columns include the lagged dependent variable of the Socialist vote.Columns 3 and 4 include time dummies, and Columns 5 and 6 add region dummies.Controlling for previous level of electoral support and unionization level, a higher level of unionization boosted the Socialist vote in the next electoral period.More precisely, the full model of Column 6 shows that among two localities that had the same unionization level two elections before the one that we are interested in, in the locality with a 1 per cent higher unionization rate in the immediately previous election, the Socialist vote was 2 percentage points higher in the current election. 18

Income and Vote in 1960
We cap our analysis by examining the relationship between the SAP and income using the Swedish National Election Study conducted in 1960 (Särlvik 1960)a time when the SAP had peaked in electoral support.Table 2 reports the proportion of SAP voters over total eligible individuals by income, for the whole country, urban districts (without Göteborg, Malmö and Stockholm so as to facilitate a comparison with our previous results for local elections) and rural districts.We also report the income distribution percentiles of each income segment.We observe persisting differences across income groups.However, that gradient was much less steep than the one at the beginning of the twentieth century.By 1960, the SAP could count on the support of almost half of the vote of the bottom quintile, particularly in urban areas, more than three times its support in the 1910s.Vote for the SAP peaked among voters in the third income quartile (Kr.10,000-14,000), falling to about 20 per cent among the top decile. 1916 We do not include results for the service sector because there was wide variance in social status (from domestic service to high professionals) in that sector.Indeed, our estimates come with wide confidence intervals.17 Again, we avoid fixed effects so as not to bias our non-linear models.
18 None of the generalized variance inflation factor (VIF) of the variables is above 5, pointing to the absence of multicollinearity in our models.

19
A logit model in which the vote for the SAP is regressed on income and income squared confirms that support for the SAP was relatively similar among low-and middle-income sectors, and peaked among individuals making Kr.10,000 (60th percentile).Notes: The election years in the analysis are 1911, 1914 (spring), 1914 (autumn), 1917, 1920, 1921, 1924, 1928, 1932, 1936, 1940 and 1944.Standard errors in parentheses.*** p < 0.01.

Figure 1 .
Figure 1.Distribution of Swedish registered voters by income bracket in 1914.

Figure 2 .
Figure 2. Proportion of registered electors that were eligible and that voted in local elections 1910, 1912-14 and 1916-18 (raw data).

Figure 3 .
Figure 3. Proportion of eligible voters abstaining and voting for the Liberals, Conservatives and Socialists in local elections by income (years: 1910, 1912-14 and 1916-1918)  (raw data).

Figures 5
Figures 5 and 6 are based on models estimated without län or year dummies so that we do not focus on a particular year or län in the interpretation of the results.14 This positive correlation between union density and abstention disappears when we include time, type (urban/rural) and län dummies.
the temperance covariate and disappears when we run the same model without logging income (see Online Appendix E).15The Consolidation of the SAPEconomic Sectors and Socialist VoteAfter 1920, Sweden's electoral authorities stopped collecting partisan vote data by any kind of demographic category after the abolition of plural voting.That leaves us with two kinds of

Figure 4 .
Figure 4. Predicted probability of participating and of voting SAP by taxable income (mean union density) and type of district.

Table 1 .
Unions and the SAP at the municipal level, 1917-44 Dependent variable: Socialist vote share (time t)