Introduction
Noting the powerful causal effect of compulsory voting (CV) legislation on election turnout, Arend Lijphart famously advocated its adoption because it would also produce a more socioeconomically equal turnout and consequently political representation as well (Lijphart, Reference Lijphart1997). Many studies have since then examined the sequences in this ‘Lijphartian argument’ whereby a near-universal turnout induced by CV attenuates the class bias. Still, the issue is far from settled and a ‘revisionist argument’ whose causal narrative refutes any such effects of CV challenges this. Judging between these two alternative theories, this paper attempts to tackle the central puzzle: Does CV attenuate class bias in turnout and political influence?
The effect of CV on various theorized dimensions of the Lijphartian argument has most often been examined by cross-national comparative research using aggregate data. This has produced analyses focusing on dimensions such as turnout rates (for reviews, see Cancela and Geys, Reference Cancela and Geys2016; Stockemer, Reference Stockemer2017), invalid voting (Power and Garand, Reference Power and Garand2007; Kouba and Lysek, Reference Kouba and Lysek2016; Martinez i Coma and Werner, Reference Martinez i Coma and Werner2019), partisan advantage (Jensen and Spoon, Reference Jensen and Spoon2011), policy outputs (Chong and Olivera, Reference Chong and Olivera2008), or measures of income inequality (Chong and Olivera, Reference Chong and Olivera2008; Birch, Reference Birch2009a; Carey and Horiuchi, Reference Carey and Horiuchi2017; Maroto and Došek, Reference Maroto and Došek2018). Yet, their results are affected by methodological challenges that include confounding variables, reverse causation, and model misspecification (Fowler, Reference Fowler2013: 161). Individual-level studies have instead approximated the effects on voter behavior of CV by counterfactual survey questions (Hooghe and Pelleriaux, Reference Hooghe and Pelleriaux1998; Mackerras and McAllister, Reference Mackerras and McAllister1999; Selb and Lachat, Reference Selb and Lachat2009), by combining survey and comparative data in multilevel models (Quintelier et al., Reference Quintelier, Hooghe and Marien2011; Singh, Reference Singh2011, Reference Singh2015; Dassonneville et al., Reference Dassonneville, Hooghe and Miller2017; Guntermann et al., Reference Guntermann, Dassonneville and Miller2020), or by leveraging age cutoff points in voting obligation using regression discontinuity designs (Jaitman, Reference Jaitman2013; Cepaluni and Hidalgo, Reference Cepaluni and Hidalgo2016; Leon and Rizzi, Reference Leon and Rizzi2016).
The final strand of research exploits the sequential adoption of CV in subnational units of single countries, allowing to estimate with greater certainty the effects of CV. These are rare opportunities because only a handful of countries provide any subnational variation in CV adoption. Only three such contexts have been identified so far: post-1945 Austria (Hirczy, Reference Hirczy1994; Ferwerda, Reference Ferwerda2014; Hoffman et al., Reference Hoffman, León and Lombardi2017; Gäbler et al., Reference Gäbler, Potrafke and Roesel2020), Switzerland (Singh, Reference Singh2015; Bechtel et al., Reference Bechtel, Hangartner and Schmid2016), and Australia (Fowler, Reference Fowler2013). This paper follows this last research tradition and examines a fourth case in which CV was adopted in some subnational units but not others – the Cisleithanian part of Austria-Hungary in 1907 and 1911. Empirically, the paper builds upon two unique – and fortuitous – circumstances of Austro-Hungarian electoral organization and reporting of its election statistics. First, comparable data from earlier election waves allows to leverage a difference-in-differences (DiD) approach to estimating the causal effect of CV adoption on some dependent variables. This is the hallmark of studies studying the subnational adoption of CV (Fowler, Reference Fowler2013; Ferwerda, Reference Ferwerda2014; Bechtel et al., Reference Bechtel, Hangartner and Schmid2016; Hoffman et al., Reference Hoffman, León and Lombardi2017; Gäbler et al., Reference Gäbler, Potrafke and Roesel2020). As the DiD design presupposes a parallel trends assumption (that trends in the dependent variable in the control and treatment groups would be the same in the absence of the treatment), other research on the effects of CV has utilized the synthetic control method that relaxes this assumption (Carey and Horiuchi, Reference Carey and Horiuchi2017; Bechtel et al., Reference Bechtel, Hangartner and Schmid2018; Singh, Reference Singh2019b; Feitosa et al., Reference Feitosa, Blais and Dassonneville2020). Second, the Austro-Hungarian statistical office reported turnout also by occupational categories in all 17 kingdoms and lands in the 1907 election, thus providing figures there that are usually unavailable in modern-day elections. These figures form the basis of inferring the extent of class bias attenuation attributable to CV.
The structure of the article is as follows. The first section reviews the state of the art on the evidence concerning the Lijphartian and the revisionist arguments. The second section delves into the historical specificities of the process whereby CV was adopted in some, but not all, crownlands. The third section examines the partisan and turnout effects of CV using the DiD framework. The fourth section evaluates the extent of class bias attributable to CV adoption. The conclusion discusses the principal findings that show little support for the Lijphartian argument as CV adoption did not result in any discernible partisan effects and did not cause the underprivileged classes to vote in higher numbers than their more fortunate compatriots.
Compulsory voting and unequal representation
The institutional prescription for CV to increase turnout is not impressive, unless accompanied by its logical conclusion that near-universal turnout induced by CV also means socioeconomically equal turnout. CV was famously advocated in the APSA presidential address as the most important solution to such class biases by Arend Lijphart for whom ‘unequal participation spells unequal influence – a major dilemma for representative democracy’ (Lijphart, Reference Lijphart1997: 1). Turnout under voluntary voting is supposed to be heavily skewed as more resourceful – richer and more educated – citizens are more likely to vote. As a result, views that contribute to policy-making are also skewed toward the rich, further widening the wealth gap (Birch, Reference Birch2009b: 23). The class bias disadvantages the lower classes whose support for policies of fiscal redistribution is not voiced in the political arena, so public expenditures are higher under CV (Chong and Olivera, Reference Chong and Olivera2008). As A. Lijphart put it: ‘(…) who votes, and who doesn’t, has important consequences for who gets elected and for the content of public policies.’ (Lijphart, Reference Lijphart1997: 4).
The argument can be summarized in three steps, all three of which constitute testable empirical propositions. Voluntary voting (as opposed to CV) results in (1) a class-biased turnout, (2) skews political representation in favor of the more privileged, and (3) generates policies that reproduce inequality. There is abundant evidence to support this ‘Lijphartian argument’. There is evidence that CV attenuates class biases by equalizing turnout rates across individual skills and education categories (Hooghe and Pelleriaux, Reference Hooghe and Pelleriaux1998; Jaitman, Reference Jaitman2013; Dassonneville et al., Reference Dassonneville, Hooghe and Miller2017). A test that included the largest number of observations in a cross-national comparison found that CV strongly irons out turnout disparities across age, political knowledge, income, efficacy, and party identification (Singh, Reference Singh2015).
There is also evidence to support the conclusion that CV bestows an electoral advantage on parties pushing for policies to reduce inequality (Leon and Rizzi, Reference Leon and Rizzi2016). An Australian survey found that those who would not turn out if voting were voluntary are more likely to vote for the Labor party, suggesting a partisan advantage of CV for the left (Mackerras and McAllister, Reference Mackerras and McAllister1999). Similarly, the experience from gradual adoption of CV in Australian states detected a strong positive effect of CV of about 9% points on the Labor Party vote share using a DiD method (Fowler, Reference Fowler2013: 172). Finally, as CV significantly mobilizes those at the bottom of income distribution, this translates into an increase of support for leftist public policies in national referenda (Bechtel et al., Reference Bechtel, Hangartner and Schmid2016). Cross-national comparisons provide indirect evidence that such public policies may indeed be effective, as CV is associated with lower levels of income inequality measured by the Gini coefficient (Chong and Olivera, Reference Chong and Olivera2008; Birch, Reference Birch2009a; Carey and Horiuchi, Reference Carey and Horiuchi2017).
The revisionist argument
However, a growing body of literature provides evidence from which a ‘revisionist argument’ can be constructed that fails to trace the causal steps in the original proposal. This revisionist argument refutes the causal effect of CV on more equal representation. The class bias attenuation hypothesis does not seem to be well supported, with some studies even presenting contradictory results (Cepaluni and Hidalgo, Reference Cepaluni and Hidalgo2016). An international multilevel comparison from 36 countries found that CV does not disproportionately induce the less educated to vote, because it also raises turnout among the more educated by the same rate (Quintelier et al., Reference Quintelier, Hooghe and Marien2011). High turnout – the mechanism through which Lijphart thought CV dissolves class biases – does not appear to operate in an international multilevel comparison which finds that the poor are no better represented in CV systems when turnout is high (Guntermann et al., Reference Guntermann, Dassonneville and Miller2020). CV does not affect the likelihood of the rich having higher turnout relative to the poor in another international comparison (Kasara and Suryanarayan, Reference Kasara and Suryanarayan2015). Experience with a recent CV abolition in Chile suggests that while CV did not generate any class biases, switching to voluntary voting introduced pro-rich biases in urban districts and pro-poor biases in the rural ones (Brieba and Bunker, Reference Brieba and Bunker2019).
The link between CV and the electoral advantage of left-leaning parties was similarly challenged on empirical grounds as the abolition of CV in the Netherlands actually helped the leftist Social Democrats (Miller and Dassonneville, Reference Miller and Dassonneville2016), and potentially abolishing it in Belgium would not greatly affect the party choice (Selb and Lachat, Reference Selb and Lachat2009). No major shifts in support between main political parties based on whether voting was compulsory or voluntary were reported using subnational variation in the 1990s in Austria (Ferwerda, Reference Ferwerda2014), and a cross-country comparison provides mixed results as CV rules significantly decrease the share of votes for the left, and CV only helps the left parties when conditioned by high turnout rates (Jensen and Spoon, Reference Jensen and Spoon2011).
As for the link between CV and policy effects favoring more redistribution, a study on the composition and amount of public spending across Austrian states reports no effect of CV (Hoffman et al., Reference Hoffman, León and Lombardi2017). As a consequence of the elusive connection between CV and policy outcomes, the overall income inequality may not be reduced by it as reported in an international comparison (Maroto and Došek, Reference Maroto and Došek2018).
The revisionist argument does not dispute the link between CV and higher turnout, which is well supported, but it challenges the notion that adopting CV translates into a less class-biased turnout, advantages parties pushing for redistributive policies, and consequently reduces income inequality. This process might be hindered by a number of obstacles. The following four, or their combinations, have been identified by the existing literature:
First, while CV may indeed foster turnout, it may also have unintended consequences on which groups are more likely to participate depending on the nature of fines. A Brazilian regression discontinuity design using age cutoff points reports that CV increases turnout of those with higher education attainment (and hence higher SES) more strongly than of the less educated (Cepaluni and Hidalgo, Reference Cepaluni and Hidalgo2016). As CV fuels political inequality rather than attenuates it, this directly contradicts the Lijphartian argument. The use of nonmonetary fines that sanction abstainers by denying their access to public services (e.g. working in the government, enrolling in a public university) affects disproportionately more those with high SES, so these turn out in larger numbers.
Second, the Lijphartian argument assumes that the additional voters drawn to the polls by CV actually cast their ballots for a specific party/candidate. But CV does not only substantially increase turnout, but also the casting of invalid (blank/null) votes which do not decide who gets to be elected (Power and Garand, Reference Power and Garand2007; Kouba and Lysek, Reference Kouba and Lysek2016; Martinez i Coma and Werner, Reference Martinez i Coma and Werner2019). As this countercurrent of CV lacks any representative consequences, depending on the share of invalid ballots cast, it may substantially reduce or even erase any effect that CV exercises on turnout (Kouba and Mysicka, Reference Kouba and Mysicka2019). This is troubling for theories that rely on the representative consequences of CV. It could mean that the class bias of the overall turnout under CV is actually attenuated, but not the class bias in valid votes. As a result, partisan effects attributable to CV may be reduced or outright eliminated by casting of invalid ballots. Research suggests that alienation and disinterest disproportionately fuel invalid voting under CV rules (Cohen, Reference Cohen2018) as the politically unknowledgeable, uninterested, untrusting, and disaffected individuals cast an invalid ballot rather than abstain to avoid sanctions (Singh, Reference Singh2019a). Furthermore, many invalid votes are cast unintentionally by error and do not appear in self-reported surveys (Kouba and Lysek, Reference Kouba and Lysek2019). Both these processes are strongly related to education levels or skills and may effectively disenfranchise the underprivileged. As CV increases the propensity of such unsophisticated voters to cast an invalid ballot, especially in party systems with many policy offerings, this raises ‘serious questions for the efficiency of compulsory voting as an institutional remedy’ for unequal representation (Moral, Reference Moral2016: 742).
Third, even if those additional voters compelled involuntarily to the polls by CV cast a valid vote, this is much less likely to reflect their preferences. Cross-national comparative survey evidence suggests that voters under CV are more likely to cast a haphazard vote that does not align with their ideology, their feelings about the parties, and their partisan attachments (Singh, Reference Singh2016). Furthermore, because CV makes these additional voters less predictable and less consistent in their vote choice, they are likely to vote in higher numbers for smaller parties at the expense of larger ones, fueling party system fragmentation (Selb and Lachat, Reference Selb and Lachat2009; Jensen and Spoon, Reference Jensen and Spoon2011: 703). This might further hinder the translation of voter preferences into policies. The Lijphartian argument assumes that voter preferences are represented by parties, and that especially those preferences regarding redistribution policies and income inequality are amplified within the party system due to CV. But if these compelled voters fail to reflect their own preferences by their vote, these presumed representative consequences may not materialize.
Fourth, the Lijphartian argument assumes that parties respond programmatically to their voters. In the context of CV, this means that parties boosted by the electoral participation of low-income voters would advance such policies that benefit these voters. But a large part of linkages between voters and parties is not programmatic, but clientelistic (Kitschelt, Reference Kitschelt2000). On the one hand, parties operating under CV were shown to rely less on clientelistic vote-seeking strategies and more on programmatic ones (Singh, Reference Singh2019b). On the other hand, there is also evidence, that one particular clientelistic strategy – vote buying which rewards voters for switching their vote choice – is actually fostered by CV because CV increases the costs of abstaining and the number of cheap vote-buying targets (Gans-Morse et al., Reference Gans-Morse, Mazzuca and Nichter2014; León, Reference León2017). This has consequences for the Lijphartian argument because pervasive clientelism helps explain why CV fails to reduce income inequality or increase social spending levels in an international comparison (Maroto and Došek, Reference Maroto and Došek2018). Despite compelling low-income voters to the polls, CV does not help to make them support redistributive policies because their votes do not follow the basic programmatic linkage of democratic accountability.
The adoption of compulsory voting in Cisleithania
Whether the Austro-Hungarian experience with CV conforms better to the Lijphartian or the revisionist argument is an empirical question. The 1907 adoption of CV there occurred in the context of major electoral reform that also introduced equal (for men) suffrage in elections to the Imperial Council (Reichsrat), a parliamentary body with deputies elected in all the 17 Cisleithanian kingdoms and lands (Boyer, Reference Boyer2013). CV was used in two elections in 1907 and 1911, the last two held before the demise of the empire in 1918. It emerged as a controversial and politically charged issue already in September 1906, when Josef Schlegel, deputy of the Austrian Christian Social Party, presented a bill to the election reform committee of the Imperial Council. It proposed that the decision of whether to introduce CV in elections to the Imperial Council be left upon the individual land assemblies (Landtag) of each of the 17 crownlands that would also lay down the necessary implementing provisions. This was already a compromising political strategy by the Christian Social Party, the main force behind CV, calculating that a wholesale adoption of CV on the whole of the Cisleithanian territory would have little chance of success and would complicate the adoption of the whole electoral reform (Kleinwächter, Reference Kleinwächter1908). In this way, the Christian Socials could secure its adoption in those lands where the assembly was under their control or that of their political allies. Following long and passionate debates between both the supporters and opponents of CV in the parliamentary committee, the parliamentary plenary as well as in the newspapers, Schlegel’s proposal, with few modifications, was eventually approved by the Imperial Council as part of the whole electoral law on 26 January 1907 (Triepel, Reference Triepel1911).
In order to implement the CV law in time before the scheduled May 1907 election, the land assemblies had to act quickly if they wished to adopt CV. The following six lands did so within a span of just 2 months: Lower Austria (which included the capital city Vienna) was the first to pass the law on February 13, followed by Silesia (March 30), Vorarlberg (April 4), Moravia (April 6), Upper Austria (April 11), and Salzburg (April 11). The remaining 11 Cisleithanian lands left their citizens’ voting voluntary for the 1907 Imperial Council election. Two more lands, however, adopted CV after this election – Carniola (20 June, 1910) and Bukovina (February 11, 1911) – so eight lands and kingdoms operated under CV and nine under voluntary voting in 1911 (Regulation no. 62, 1911). Sanctions for not voting consisted of fines set from 1 to 50 crowns (Kronen), which the abstainers were obliged to pay unless they presented a ‘justifiable excuse’. Such possible excuses varied somewhat between lands. Being at least 25 kms distant from one’s voting poll on the election day constituted an excuse in Moravia, while only those who traveled to another land were excused in the remaining lands and kingdoms. Those over 70 years of age were excused in Silesia and Moravia but not elsewhere. The laws also excused those who were ill or those who were kept from voting by urgent family or occupational exigencies (Regulation no. 62, 1911).
In the central parliament as well as in the land assemblies, it was the Christian Social Party that pushed for CV. Lacking any historical precedent in the empire, Schlegel’s proposal was rationalized by precedential usage of CV in Belgium and some Swiss cantons (Kleinwächter, Reference Kleinwächter1908), but also by normative arguments, with the positive educational effect of CV on the citizenry being most frequently emphasized by CV supporters. There is little doubt, however, that the push for CV by the Christian Socials was a thinly veiled strategy to gain an electoral advantage over their main rivals, the Social Democrats. The Christian Social Party in Austria was a coalition of urban bourgeois and rural middle peasants who dominated this alliance due to the exceptional cohesiveness of the middle peasantry (Luebbert, Reference Luebbert1987: 476). Their insistence on CV must be viewed in a context of their striving to build a mass party to gain an electoral advantage – it was a ‘(…) a crude, but effective, tactic to force the Austrian bourgeoisie into successful competition with the Social Democrats.’ (Boyer, Reference Boyer1986: 170). Albert Gessmann, the main architect of CV adoption and the Christian Social Party parliamentary leader, summarized this point succinctly during the election reform committee debate: ‘The crux of the issue of compulsory voting is the terrorism exercised by the Social Democratic Party.’ (Stenographisches Protokoll, 1906: 566). The prevailing belief among the bourgeoise parties, Gessman included, was that the Social Democrats wielded tightly organized parties with disciplined and easily mobilizable followings. Such a party would gain little from CV because it had already reached the ceiling of its potential electorate. Conversely, the middle-class and rural parties with much looser organizations and undisciplined voters would disproportionately benefit from CV adoption.
The Social Democrats mobilized against the adoption of CV. In 1 October, 1906 meeting of the constitutional committee, the chairman of the Austrian Social Democrats Viktor Adler presented the party’s main objections. CV is a ‘reactionary tendency’, by which he meant ‘(…) the tendency to emphasize the weight of those strata of the population, which think politically the least of all, and have no actual political will.’ (Stenographisches Protokoll, 1906: 562–563). As a result of low education or simply of the social milieu in which they live, these socially disadvantaged (deklassiert) segments of the society lack any political interest whatsoever. CV resting on police violence would place them at an advantage over those who think politically. The causal process of CV adoption conforms easily to an interest-based account of CV adoption in which parties of the right are expected to push for CV when they are faced with a strengthening left (Helmke and Meguid, Reference Helmke and Meguid2007). CV in Cisleithania was expected to bolster the electoral fortunes of parties with rural, conservative, agrarian, and middle-class electorates; and punish parties resting their support on the industrial proletariat.
Estimating the effects of CV adoption on turnout and voting patterns
Following the existing subnational research on CV, the DiD estimator is used, in which the changes in the dependent variable in units that implemented CV are compared with simultaneous changes in units that retained voluntary voting. Assuming parallel trends in the average values of dependent variables in the absence of treatment for both the treatment and control groups, the coefficient α constitutes the estimate of the causal effect of CV on the dependent variable, estimating this equation using OLS:
$$DV = \alpha {{\rm{(}}Compulsory\,voting{\rm{)}}_{it}} + {\gamma _i} + {\delta _t} + {\varepsilon _{it}}{\rm{.}}$$
The DiD approach is used separately to estimate the effect of CV on six dependent variables: turnout, the vote shares of the Social Democrats and Clerical parties, and three measures of party system fragmentation. It includes a dummy variable for CV, land fixed effects (γ i ), and election wave fixed effects (δ t ).
As discussed earlier, the adoption of CV was endogenous to the political strategies of the principal political parties. Such endogeneity could bias results in a regression framework. The DiD approach is not affected by this problem, because the policy intervention does not need to be exogenous to land-year-specific shocks (Abadie, Reference Abadie2005: 2–3; Bechtel et al., Reference Bechtel, Hangartner and Schmid2016: 756). The DiD approach assumes that CV is uncorrelated with unobserved time-varying unit characteristics once the model controls for time-invariant, unit-specific factors (Hoffman et al., Reference Hoffman, León and Lombardi2017). If, for example, crownlands dominated politically by the Christian Social Party were more likely to adopt CV, this would be absorbed by the land fixed effects. The descriptive statistics for the variables in the following analyses are listed in Table 1.
Table 1. Descriptive statistics
Only 2 elections were held in some of the 17 Cisleithanian crownlands under CV – in 1907 and in 1911, which was also the last election before the empire’s demise in 1918. Both elections were governed by two-round majoritarian systems with a closed runoff for the two biggest vote-getters. A total of 516 deputies were elected in 480 districts. In 16 of the 17 lands and kingdoms, these constituted exclusively single-member districts. The runoff had to be held in 215 districts in 1907 and 208 districts in 1911 (Riemer and Klezl, Reference Riemer and Klezl1911: 673). Only two lands, Bukovina and Carniola, switched to CV between 1907 and 1911, so the DiD estimation would only employ observations on change of CV rules from these two units if solely the 1907 and 1911 data were used. However, the DiD approach requires a wider set of observations where a switch to CV occurred. Fortunately, data is available for the preceding two Imperial Council elections held in 1897 and 1900/1901 under voluntary voting in all 17 crownlands. These elections were the last ones to use the curial system, which divided voters into five curiae (large landowners, urban, rural, chambers of commerce, and general). As a consequence, using the overall turnout rate is not a possibility as a comparison group to later equal and universal franchise, because the curiae had, of course, a highly skewed socioeconomic composition.
Nevertheless, both the 1897 and 1900/1901 elections included for the first time the fifth, general, curia following the 1896 election reform pushed for by Prime Minister Kasimir Badeni. In this curia, suffrage was universal (for men) and equal for those older than 24 years. Although 72 out of the total of 425 deputies for the Imperial Council were elected in the 17 crownlands in the general curia in 1897, this inclusion meant an immense expansion of the franchise from 1.7 million to 5.3 million eligible voters in the 1897 election (Hloušek, Reference Hloušek, Balík, Hloušek, Kopeček, Holzer, Pšeja and Roberts2017: 24). This figure roughly corresponds to the 5.5 million eligible voters in the 1907 election which eliminated the curial system. Importantly for comparison with the 1907 and 1911 elections, the system in 1897 and 1900/1901 retained plural suffrage, meaning that all those eligible to vote in any of the four previous curiae were also eligible to vote in the general curia, increasing the weight of their votes (Beneš, Reference Beneš2017: 106). Because the 1897 and 1900/1901 elections had within the general curia a universal and equal suffrage for the identically defined (and similarly sized) electorate it is, therefore, the 1897 and 1900/1901 data on the respective variables from this general curia that enters the DiD model together with information about the 1907 and 1911 elections. Any time-invariant features of these two previous waves of elections, such as the higher number of voters per deputy or the context of the curial system, are controlled for by virtue of the time-fixed effects in the DiD. In sum, data on 68 observations from the 17 crownlands in 4 election waves enter the DiD estimation, exploiting a switch in the value of the independent variable (voting compulsion) in the 6 lands that adopted it in 1907 (plus the 2 that did so in 1911). All analyses are based on data on electoral, census, and occupational characteristics published in their respective volumes by the Austro-Hungarian Central Statistical Office (Österreichische Statistik, 1902, 1908, 1912a, b, 1914). Estimates of the GDP per capita in 1900 were obtained from Good (Reference Good1994).
The DiD design is presupposed on the parallel trends assumption that the outcome variable would on average have parallel trends across units in the absence of treatment. One way of ascertaining its plausibility is to focus on the pre-treatment trends in the dependent variable. Figure 1 divides the crownlands into three groups and plots the average turnout rate in each group across the four elections: (1) those that retained voluntary voting in all four elections, (2) those that adopted CV in 1907, and (3) those that adopted CV in 1911. Note that the last group only contains two crownlands. The variation in pre-treatment turnout suggests the DiD estimation is not invalidated, as turnout declined parallelly in both the first and second groups between 1897 and 1900/1901. Turnout remained low in the general curia with universal male suffrage in those two elections (which is attributable to its limited voting power relative to the other curiae), but substantially increased in 1907.
Figure 1. Average turnout in three groups of crownlands according to the year of CV adoption (1897–1911).
Table 2 presents the DiD estimates. It presents models without any control variables. When important sociodemographic characteristics (proportion of Catholics, GDP per capita, and illiterate population) are controlled for, the principal conclusions remain the same (results not presented). The first dependent variable is the turnout rate in the decisive election round. This includes turnout rates either in the first round if it resulted in a candidate victory, or in the runoff if that was not the case. The contribution of CV to turnout is large – an estimated 13.9% points. This is lower than the effect of adopting CV in Australian states estimated at 24.3 percentage points (Fowler, Reference Fowler2013: 171), but about double the size of CV adoption for parliamentary turnout (6.6 percentage points) in post-1945 Austrian states (Hoffman et al., Reference Hoffman, León and Lombardi2017).
Table 2. Difference-in-differences estimates of the effects of compulsory voting, 1897–1911
* P < 0.01.
Standard errors in parentheses, SE.
Was CV effective in reducing the Social Democratic threat, as hoped for by its architects, or was it rather a tool for strengthening the political representation of the working classes by advantaging the Social Democrats as predicted by the Lijphartian argument? The Social Democratic vote share in the 68 crownland-years was calculated from election statistics. The DiD approach does not find support for either of the two predictions. The coefficient on CV is substantively small and not significant. CV did not bestow an electoral advantage on the Social Democrats, but it did not disadvantage them either.
The Clerical parties’ voting strength was similarly unaffected by CV. The Clerical party family was more fragmented than the Social Democrats and evolved differently in different lands. In order to facilitate comparisons over time, the vote shares of all clerical, Catholic, and Christian Social parties were summed up in each land in each of the four elections to create this dependent variable.
On the other hand, it is possible that the partisan effects of CV are channeled through other parties than either the Social Democrats or the Clericals. Despite these two forces being pivotal in most Cisleithanian lands, a plethora of other parties also competed in the elections. In other words, the findings do not rule out the possibility that CV (dis)advantages some parties over others. Consequently, it may be the case that the surge in turnout attributable to CV of some class-based constituencies benefits some other party. Because the identity of most other parties is specific to each land, it is not possible to directly test this hypothesis in a way similar to the DiD approach toward the Social Democrat and Clerical party families. Indirectly, however, one might infer that if CV benefits marginal parties as proposed in the theoretical section, this would translate into an overall higher fragmentation of the party system. To check for this possible effect, party system fragmentation was ascertained in all the 68 crownland-years and examined using the DiD approach described above. Four variables were calculated to measure party system fragmentation: (1) index of the effective number of parties (which divides one by the sum of squared shares of votes for each party in every crownland), (2) the raw count of parties, (3) the biggest vote-getting party share (in % of the vote), and (4) the combined share of the two biggest vote-getters (in %) to consider party dominance. The results in table 2 suggest that party system fragmentation in the crownlands was not influenced by whether CV was adopted or not as CV adoption did not significantly impact any of the four outcomes. This represents indirect evidence strengthening the conclusion that CV produced no discernible partisan effects.
In spite of the powerful positive effect on turnout rates attributable to CV, this increase did not electorally benefit the main partisan contenders in Cisleithanian politics. This is unpredicted by the Lijphartian argument. The next section explores some reasons why that could have been the case.
The effect of CV on class bias in turnout
Following the Lijphartian argument, even in the absence of direct partisan effects, CV could still contribute to a more socioeconomically equal political representation by emphasizing the political influence of lower strata of the society over the influence of the society’s more privileged members. To disentangle this problem, it is necessary to examine how different classes would participate in elections if they were exposed to voting obligation as opposed to voluntary voting. Breakdown of turnout statistics by occupational (or other) groups is usually unavailable in modern electoral statistics. The attractive feature of the Austro-Hungarian statistical reports is that the count of eligible voters and the count of actual voters was recorded for all 17 crownlands and for identically defined 35 occupational categories in the 1907 election (Österreichische Statistik, 1912b). Many of these occupational categories are too narrow and/or too small, but the following analysis aggregates these into conceptually distinct classes. Of course, occupational categories do not exhaust class stratification typologies, as classes are a multidimensional phenomenon. But occupation (usually together with education) is the primary dimension differentiating among social classes understood as ‘aggregates of individuals who occupy broadly similar positions in a hierarchy of power, privilege and prestige’ (Williams, Reference Williams1960: 98; Kohn and Schooler, Reference Kohn and Schooler1969: 660).
Table 3 proposes such class stratification of the Cisleithanian electorate divided into three principal groups (working class, lower middle class, and middle class) and their main constituent categories. Their relative size in the total number of the 5,464,180 eligible voters is presented in the first column. One potential problem would arise if these occupational categories lacked any class-based connotations and political relevance. Because the connection between class status and vote choice cannot be reconstructed with full certainty (as the vote was secret and surveys nonexistent in 1907), it must be inferred indirectly.
Table 3. Turnout of social classes in crownlands with and without CV in 1907
The sizes of individual categories do not add up to 100%, but only to 95.8% as occupations of the remaining 4.2% eligible voters were not recorded or the recorded categories are vaguely defined.
To check for the validity of this assumption within the industrial working classes, it is possible to relate their relative strength in all crownlands to the electoral strength of the Social Democrats. The data suggests that voting for the Social Democrats was heavily conditioned by class structures corresponding to these occupational categories (see Figure 2). There is a very strong correlation between the proportion of industrial workers in both large and small enterprises (on the total electorate) to social-democratic vote share (Pearson r = 0.88 in 1907). The relative strength of occupational groups serves as a strong predictor of voting behavior.
Figure 2. Relationship between vote share for Social Democrats and the share of industrial proletariat in the 1907 elections to the Imperial Council.
It is difficult to validate in the same direct way the relationship of other class-based constituencies to voting patterns, because their vote was often divided across ethnic and confessional lines for parties that were idiosyncratic to the individual crownlands. For example, while the ethnically German peasants voted for the Christian Social Party in most crownlands, Czech peasantry in Bohemia voted for the Agrarians who disagreed over the role of the Catholic Church, while the Polish peasants split their vote between liberal, Catholic, and radical agrarians (Jenks, Reference Jenks1967: 58). Furthermore, a significant and politically consequential cleavage characterized class composition in the rural areas differentiating between the agrarian proletariat and independent middle peasantry (Luebbert, Reference Luebbert1987). The category of landless peasants reported by the election statistics includes landless laborers, but also the more varied category of rural house owners with little or no land (Häusler), conforming to a definition of agrarian proletariat as landless workers and smallholders dependent on the agrarian labor market for some part of their income (Luebbert, Reference Luebbert1987: 462).
While it is perhaps elusive to find a stringent definition of the ‘lower-middle-classes’ or the ‘petite bourgeoisie’, it could be affirmed that the core of its identity rests on the professional autonomy enjoyed by the independent smallholding peasants, craftsmen, shopkeepers, artisans, independent professionals, peddlers, or small traders (see Mayer, Reference Mayer1975; Scott, Reference Scott2014). Politically, these ascendant lower middle-class groups asserted themselves as insurgent movements after the mid-1880s challenging the older liberal middle-class parties in Cisleithania, and engaging through a multitude of parties in ‘the aggressive advocacy of petit bourgeois social interests’ (Cohen, Reference Cohen1998: 51). At the very least, this heterogeneous group should be divided based on agrarian and non-agrarian lines. The final distinction concerns the sizable group of craftsmen distinguished from other independent professions including categories of self-employed professionals in transport, business, and free occupations (freie Berufe).
Within the last class category, the Austrian officials (Beamten) constituted a clearly delineated middle-class group deriving its status from patronage provided by Mittelstand parties and wielding special weight and influence in public affairs (Boyer, Reference Boyer2013: 152). Separate from, but close to, the aristocratic upper class, the empire’s bureaucracy (Beamtentum) remained closely intertwined with its bourgeoisie (Barany, Reference Barany1998: 209). The reported occupational group of officials (Beamten) aggregates 10 categories of officials including managers and administrators in industrial, agricultural, and service enterprises, civil servants, and other white-collar professionals.
The last three columns in Table 3 are relevant, if only indicative, for assessing the Lijphartian thesis of class bias attenuation. They provide turnout rates (actual over eligible voters) in each category separately for those living in crownlands with and without CV. In total, 84.6% of Cisleithanian voters turned out in 1907, but only 81.2% in the 11 lands with voluntary voting and a full 93.1% in the 6 lands using CV. The total difference between these rates in 1907 (11.8%) is close to the DiD estimate of 13.9% from all election waves. The last column examines whether distinct social classes turned out at different rates in the Cisleithanian crownlands with and without CV by subtracting turnout in Cisleithanian territory under voluntary voting from turnout in territory under CV. (Note that these figures are not averages across crownlands.) Considering that these are only descriptive figures, the data suggests little support for the Lijphartian argument. According to that theory, CV was expected to cause the largest differences in turnout in the categories of underprivileged citizens. But the descriptive figures suggest that the differences in turnout rates across the three main class categories are essentially the same. Furthermore, one particular category within the working classes – the industrial workers in large factories – was little mobilized by voting compulsion. With only a 5.1 percentage points contribution to its turnout, this contradicts the Lijphartian argument. It is, on the other hand, fully consistent with the hopes of Cisleithanian CV designers and fears of the Social Democrats. The industrial proletariat was already mobilized prior to CV introduction and participated intensively in elections even without voting compulsion.
To complement these descriptive figures, models in Table 4 formally test the effect of CV on a turnout of all 10 class categories. The dependent variable is turnout within an occupational category in a crownland. The 10 occupational categories in the 17 crownlands result in a total of 170 cases. The independent variables indicate the pertinent occupational category and whether CV was adopted using dichotomous measures. The interaction terms of CV with each of the occupational category dummies evaluate the claim of differential effects of CV across different classes. They formally test the equality of coefficients of the interactions between the different occupational categories. This procedure has the additional advantage that potentially confounding variables may be controlled explicitly in the model. As eligible voters were not assigned randomly to elections with and without CV, unobserved differences between the crownlands may confound the relationship between CV and class bias in the preceding descriptive analysis of the 1907 class voting. It is possible that the absence of a class bias in the increased turnout is a result of some other features of the crownlands that adopted CV that contributed to its underprivileged classes voting in lower numbers. The model controls for 6 potential confounders that are likely to affect electoral participation in the 17 crownlands. The percentage shares of the literate population, of the largest linguistic group (as a proxy for ethnolinguistic fractionalization) and of Catholics as well as the GDP per capita, population and population density capture different dimensions of modernization, demography, education levels, and cultural values are known to affect electoral participation (Cancela and Geys, Reference Cancela and Geys2016). It is essential to note that this test only relies on the 1907 election data because information on turnout within classes is not available for the other election years. Compared to the DiD approach, this is a much weaker methodological design with less credible estimates, not least because there might be other sources of spuriousness not accounted for by the control variables.
Table 4. The effect of CV introduction on turnout across social classes in 1907
* P < 0.1, **P < 0.05, ***P < 0.01.
Standard errors in parentheses, SE.
The two models in Table 4 differ in the selected reference category of the occupational group. The industrial workers in large enterprises are the reference (i.e. omitted) category in model 1, the independent peasants are the reference category in model 2, and the landless peasants in model 3. These three groups constitute the largest groups and were also chosen for theoretical reasons. Comparing the turnout rates of the underprivileged industrial proletariat to a turnout of other classes directly tests the Lijphartian thesis. Relating the turnout rate of the independent peasants to other classes tests the effectiveness of the Christian Social assumption of gaining electoral advantage among the middle peasants, but also compares the largest single Cisleithanian occupational group (37% of all eligible voters) to all others.
The coefficients on the interaction terms capture the effect of moving from voluntary voting to CV in any given occupational category (relative to the reference occupational category) over and above any effect, we observe under voluntary voting. For example, the coefficient of −1.87 on the interaction term between CV and the independent peasants’ category means that the turnout rate of independent peasants was only 1.87% points lower than that of the industrial proletariat as a result of CV introduction. But as this is not statistically significant, we cannot rule out that CV introduction did not matter at all.
Overall, the results from both models suggest that CV introduction did not differentially affect different classes, which runs counter to the Lijphartian thesis and supports the revisionist argument. The industrial proletariat was no more or no less likely to turn out to vote as a result of CV than constituents of other lower classes, lower middle classes, and most middle classes as the insignificant coefficients suggest. The only exception is the difference in turnout rates from groups of teachers and clergy. CV resulted in about a 10% point deficit in turnout of both clergy and teachers compared to industrial workers. This is consistent with the Lijphartian argument as these highly educated professionals already voted in high numbers so voting compulsion did not affect their turnout as much as of the less privileged classes. On the other hand, their share in the society (0.7% for teachers and 0.4% for clergy among the total eligible voters) was so small that this could hardly affect electoral results. The much larger middle-class group of officials (constituting 4% of eligible voters) did not vote in lower numbers compared to industrial workers as a result of CV. Similar conclusions apply to differences in turnout from the largest single social class in Cisleithania – the middle peasantry. With a partial exception of the clergy, CV did not result in higher or lower turnout rates compared to all other occupational groups. The absence of any meaningful differential effect, especially compared to urban industrial workers, also reveals that the hopes of the Christian Social reformers who pushed for CV did not materialize. CV was not an effective tool of mobilizing their rural supporters at the expense of organized urban proletariat.
In sum, not only did CV not translate into electoral (dis)advantages for particular parties, but it also failed to spur the attenuation of differences in turnout across the social classes.
Conclusion
CV was adopted in the late Habsburg empire by 6 of the 17 Cisleithanian crownlands in 1907, followed by 2 more in 1911. Leveraging this variation, it is possible to estimate the effects of its adoption with greater certainty than is feasible in analyses where the variation in CV/non-CV rules comes from different sources. The overarching question is, after all, what CV legislation would affect if it were adopted in any given setting; and not about cross-national variation in CV rules, variation in neighboring age groups around CV cutoff points, or counterfactual questions to voters under CV. On the other hand, the lessons that could be learned from this reform are limited because only two elections were held using CV before the demise of the empire in 1918. Note, furthermore, that this occurred before the advent of public opinion polling, so parties were unable to adjust their electoral strategies to public response to CV beyond observing election results.
The conclusions are discordant with the argument forcefully advanced by A. Lijphart, which has nourished the ongoing lively debate on the merits of adopting CV. Despite data limitations due to the historical elections, the Austro-Hungarian experience with CV rather consistently conforms to the ‘revisionist argument’. Introducing CV did indeed increase turnout and this boost was substantial (about 14% points), but this is also the last step that is consistent with the Lijphartian argument. The introduction of CV in Cisleithania did not have any discernible partisan effects – it did not benefit parties representing the lower classes, smaller parties, and not even the parties that pushed for the introduction of CV. This lack of partisan consequences was most likely due to the fact that the contribution to turnout attributable to CV was spread out equally across social classes. The lower classes were no more likely to turn out due to CV than were the higher echelons of the society.
This is not the first study to note that large increases in turnout due to CV do not necessarily translate into partisan effects. The same conclusion was reported from post-1945 Austria, where the contribution of CV to turnout is likely powered by disinterested and uninformed voters lacking strong policy or partisan preferences and whose voting patterns consequently do not benefit any particular party (Hoffman et al., Reference Hoffman, León and Lombardi2017). But another mechanism was also likely to be operative in the Cisleithanian elections – the segment of voters who went to the polls due to CV did not differ much in its class composition from those who turned out even in the absence of sanctions.
Interestingly, only one half of the principal reasoning behind CV adoption emphasized during the 1906 reform debates has some support in the data. Industrial workers in large factories were the occupational group whose turnout featured the smallest gap between crownlands using voluntary voting and those using CV. The architect of CV, Albert Gessmann, anticipated this, claiming that when hundreds or thousands of people are concentrated in a factory, their organization and agitation is ‘child’s play’ (Spielerei) and therefore Social Democracy ‘has an incredibly easy job with its organization’ (Stenographisches Protokoll, 1906: 566). Because of this, its followers were already highly likely to cast their ballots even in the absence of voting compulsion, so any boost by CV would soon hit the ceiling of near-universal turnout. In Bohemia, the most industrialized crownland concentrating almost one half of all Cisleithanian large-scale industrial proletariat, the turnout within this group was an impressive 90.3% under voluntary voting. In neighboring and ‘most similar’ Moravia with CV and the empire’s second largest industrial workforce, their turnout was only barely 4 points higher (94.0%).
On the other hand, when other determinants of turnout are controlled for, the industrial proletariat exhibited no smaller turnout rate increases as a result of CV than did other classes. Furthermore, CV did not electorally punish the Social Democrats, as originally hoped for by the CV architects and feared by the Social Democrats themselves. Nor did the rural masses gain disproportionate political influence due to CV. Ironically, the causal beliefs of political elites about the effects of electoral rules on their electoral fortunes are very often not fulfilled after these rules are instituted (Andrews and Jackman, Reference Andrews and Jackman2005). Such elite miscalculations do not result from the lack of self-interested electoral rule choice but rather from the lack of information in contexts of extreme uncertainty, a situation that characterized the adoption of CV in Cisleithania.
Acknowledgments
The author acknowledges the funding provided by the Philosophical Faculty, University of Hradec Králové (grant ‘Formal and Informal Institutions and their Impact on Political Representation’ within the Internal Grant Competition). I am sincerely grateful for the helpful comments from three anonymous reviewers. Finally, I would like to thank Radek Buben and Jakub Lysek for generously commenting on earlier drafts of this article.
