The Inequality Roots of Weak Partisan Attachment: Evidence from Denmark

,


Introduction
How does economic inequality affect the nature of national party systems? While the classification and description of political parties and national party systems are among the most classic topics in political science, there has recently been a surge in the interest in national party systems and their broader impact. Strong and stable party systems, both in democracies and even autocracies, have been found to positively influence economic growth (Bizzaro et al. 2018), welfare state development (Rasmussen and Knutsen 2017) and even fiscal discipline and the (dis)use of opportunistic electoral manipulation (Hanusch and Keefer 2014).
However, in order for a country to develop a strong and stable party system, a certain level of voter attachment and loyalty to established political parties are needed (Dalton and Weldon 2007). If a large number of voters constantly shift from one party to another (including new emerging parties), it is hard to a party system consisting of the same established old parties to emerge. 1 Thus, the determinants of the development and erosion of voter loyalty to political parties are non-trivial concerns.
Previous research has tried to theorize and empirically test which factors determine the decline of partisan attachment, especially in Latin America (Roberts 2012;Lupu 2013;Lupu 2014;Novaes 2018), but much less research has been done on which factors determine the development of voter attachment to parties in the first place, 2 including what role economic and social structures play during these processes. While a growing literature has tried to investigate the determinants of electoral volatility, and thus to some extent party stability, in young and developing democracies (Tavits 2005;Tavits 2008; Feree 2010), 3 we know very little about which factors determined the level of voter attachment to political parties, and thus the stability of the party system, in the developed and old democracies of Northen and Western Europe. We know especially little about this process during the time 1 Of course, political competition in which incumbents are able to lose power is a key component of democracy and a determinant of the quality of policy-making (Besley 2006), but this does not necessarily require large swings in voter support and/or the emergence of new parties.
2 See Dinas (2014) for an important exception 3 See also Matakos and Xefteris (2016) for a study on Greece.
when the party systems of these countries developed in the first place during the late 19 th and early 20 th century. However, see Ziblatt (2017) and Charnysh and Ziblatt (2018) for new research on Germany. As voter loyalty to established parties has declined substantially in Western Europe in recent decades, and electoral volatility thus has increased (Dassonneville 2018), it is important to know what determines the development of voters' attachment and loyalty to political parties, including which factors determine whether new and emerging parties are able to build a loyal core voter base.
This article deals with how economic inequality, a potentially important but so-far understudied factor with regards to partisan attachment, 4 affects the incentive and ability for new political parties to build strong party organization and develop partisan attachment during the years of these political parties' formation and consolidation. It develops a theory of how new political parties representing rural interest are disincentivized from building strong party organizations and fermenting voter loyalty in areas characterized by high land inequality under a political system characterized by a limited franchise, a situation found in most Western and Northern European countries during the days of their early party system formation. Due to the path dependence of party organization and voter loyalty, these dynamics have wider political consequences after democratization, where areas with historical high levels of land inequality will see less electoral support for traditional agrarian parties and more support for established non-rural parties as well as new parties.
This theory is tested using difference-in-difference analysis on a dataset of election results in Danish counties in the years 1901-1943 using new and unique data on historical land inequality in Danish counties. The results show substantial evidence in favor of the theory. The agrarian Liberal Party, which emerged to represent rural interest in Denmark during the years of the limited franchise in the late 19 th and early 20 th century, did much worse electorally after democratization in areas characterized by historically high levels of land inequality, even as the Danish land reforms during the 1920s had substantially lowered land inequality. The beneficiaries of the Liberal Party's electorally weakness in areas with historically high levels of land inequality seem to have been the centrist and far-left parties as well as the extreme right. These results suggest that the effect of historically high levels of land inequality does not merely capture larger redistributive preferences in high-land-inequality areas and that a legacy of high economic inequality might leave a more permanent polarizing mark on the party system, even after inequality has markedly declined.
Apparently, the combination of economic factors, including economic inequality, and political institutions, such as a limited franchise, can have longerterm political consequences for the national party system, even long after these economic and political institutions have been reformed. These insights hold substantially implications for the study of political parties and party systems both in new and established democracies today.
2 Theory: Land inequality and the early building of rural party attachment In a society characterized by a large rural population, it is crucial for political parties who aspire to be electorally successful to win votes in rural areas. 5 Formal party organization with formalized party membership, local party offices as well as campaign experience of local party politicians help with mobilizing voters and build and maintain partisan loyalty among the electorate, 6 especially in new democracies or political systems without any previous formal political organizations in the country-side. However, the build-up of local party organization, especially in rural areas, and the build-up of a core voter support base represent a substantially amount of sunk costs. Furthermore, voters who are already mobilized/socialized to hold sympathy and to vote for a given political party are much more likely to keep supporting that party, an effect which might also spill over into these core supporters' children.
Consequently, political parties' initial strategies for party organizing and 5 In general, large political parties in democratic and even hybrid regimes need to be able to appeal to broad segments of society both in terms of social groups and geography.
6 See evidence for the role played by formal party organization for the electoral stability of left-wing parties in Bartolini and Mair (1990, 231-237). voter cultivation exhibit substantial path dependence. Thus, at the critical juncture of initial formal party formation, political parties' incentives and ability to build local party organization and a strong core voter base locally might have substantial consequences for the later electoral success of these parties. This is true, even as the initial economic and political conditions which gave rise to the political parties' initial strategies change. 7 With regards to parties with pro-rural platform, I argue that the combination of political institutions, specifically the structure of voting rights, and the level of land inequality locally during the time of a rural-based political party's initial formation strongly affects that party's incentive to build local party organization and mobilize rural inhabitants politically. Thus, the level of local land inequality during the critical juncture of rural party formation, strongly affects the later electoral fortunes of rural-based political parties, even when both political rights and land inequality have changed. Below, I flesh out this argument in greater detail.

Rural party-building under a limited franchise
Imagine a country still largely dominated by agriculture with competitive yet not fully democratic electoral institutions with the franchise limited to (male) head of households, 8 where formal political party organization is still in its infancy. An aspiring political party with a mainly pro-agricultural/prorural platform needs to decide in which areas to primarily invest time and energy in building local party organization and cultivate a core voter base. E.g. where party leaders should run as candidates, where the party should mainly attempt to publish and distributive its newspapers and where party leaders and other charismatic party orators should go on speaking tours and campaigns. 9 Given an agrarian party's ideological and policy platform, it will choose mainly rural communities as its targets for party organizing and 7 This line of argumentation builds heavily on a historical-institutional approach to studying politics (Skocpol and Pierson 2002). 8 A situation which applies to Denmark and most Western countries in the period leading up to democratization in the 19 th and early 20 th century. 9 The main type of political mobilization before the rise of modern electronic mass media.

The role of land inequality
However, since voting rights is limited to (male) head of households, the social structure of the rural/non-urban areas, especially the distribution of land ownership, the level of land inequality, also matters for agrarian parties' early party building strategies. Since farms and estates are on average larger in areas with higher land inequality, and a larger fraction of the adult (male) population is thus employed as live-in farmhands and domestic servants rather than being head of their own smaller agricultural households, rural areas with a high level of land inequality simply have fewer voters as a share of the (male) population than rural areas with lower levels of land inequality. This lowers the incentive for agrarian parties to invest much resources in political campaigns and the build-up of local party organizations in rural areas with a high level of land inequality, since there is simply fewer voters and potential party members to be had. E.g. the number of potential subscribers to party newspapers is lower so it might not be worthwhile financially to start a local party newspaper. Furthermore, areas with higher levels of land inequality might have more hierarchical social relations and norms (Ziblatt 2009;Rink and Hilbig 2018) which lowers engagement in horizontal political organizations such as local party organizations. Instead, these new agrarian political parties, whether they represent middle class interests against an older elite, or whether they are attempts for an older estate-holding elite to broaden their political base, 10 are more likely to rely on informal or weak political organizations and social hierarchies to mobilize voters rather than formal party organizations and electoral campaigns (Ziblatt 2017). See also Bartolini (2000, 482-483) for the role played by land inequality in Italy for the lack of a coherent Italian agrarian party.
Consequently, during the critical juncture of early agrarian party formation, areas with a high level of land inequality will see less formal party organisation of the main agrarian parties and will experience less campaigning and attempts at building a loyal core voter base, compared to areas with lower land inequality. Thus, fewer people (both voters and non-voters) in areas characterized by high levels of land inequality will acquire partisan loyalty to the agrarian party compared to people in areas with lower levels of land inequality. Either because they cannot simply vote for the agrarian party and develop partisan loyalty in this way (Dinas 2014) and/or because they are less connected to the main agrarian party through local party organizations, local party newspapers as well as campaigns and social events organized by the main agrarian party. Here, it is important to state that even citizens without any official voting rights at this stage could, to a varying degree, have developed partisan loyalties/attachments and been political active within local party organizations. As suggested by both historical and recent social science research, social and demographic groups (including women) often were substantially politically interested/active and had developed informal and formal ties to political parties/political organizations long before they officially gained the right to vote (Hill 1996;Teele 2018). 11

Path dependence after democratization
Eventually, the country might fully democratize and lift the previous restrictions of the franchise. In this case, political parties will now have to be able to appeal to previously non-franchised parts of the population in order to maintain and/or increase their vote share. However, if the national party system has at least partly developed under the limited franchise and to some extent has become "frozen", like in most parts of Western and Northern Europe (Lipset and Rokkan 1967), then there is a substantially level of path dependence with regards to both party organization and partisan loyalty among the electorate which have wider political consequences after democratization. 12 As mentioned, formal party organization, including the founding of local party chapters and local party newspapers, require a lot of sunk costs with regards to time, money and effort and is thus not easily established or undone in the short term. Thus, pre-democratic parties are not necessarily able to 11 See also Ziblatt (2017, 89;97-98) for the role played by non-enfranchised women in the British Conservative organization, the Primrose League, in the late 19 th century.
12 See also Charnysh and Ziblatt (2018) for a similar argument based on German data.
quickly build up stronger local party organizations in anticipation of democratization in areas where they have historically been organizationally weaker. Especially, since this requires the help of local party activists of which there might be fewer in areas with historically higher levels of land inequality due to the reasons stated in the previous section.
Additionally, partisan loyalty and identification and thus party choice are slowly changing factors which might even be passed on through generations (Dalton and Weldon 2007, 184). 13 Thus, the previous differences among rural areas with regards to partisan attachment to the main agrarian political party both among voters but especially among previous non-voters, which was a function of difference in local land inequality between these areas, cast a long "shadow" of partisan attachment well into the democratic period.
Consequently, since agrarian party organizations have historically been weaker and loyalty and attachment to the main agrarian party/parties been lower in areas with historical high levels of land inequality, the newly enfranchised voters are less mobilized and socialized to vote for the main agrarian party in areas with high historical land inequality compared to the areas with lower historical land inequality. Thus, areas with high historical land inequality should see a lower vote share to the main agrarian party/parties after democratization compared to areas with historically lower levels of land inequality. A main agrarian party should, everything else being equal, still be better able to appeal to rural voters after democratization but this rural voter "premium" will be smaller in areas with high level of historical land inequality due to the reasons stated above.
Instead, either new political parties or political parties with traditionally low rural appeal should be able to fare relatively better in rural areas with historically high levels of land inequality compared to rural areas with historically low levels of land inequality. This is true, even if the economic structure has equalized, since there is a substantial level of path dependence with regards to local party organization and partisan loyalty among the local electorate. Thus, even if land inequality decreases and even as agricultural land loses its relative importance as an economic asset, a history of high land inequality during the critical phase of initial party formation has longer-term impact on the electoral fortunes of a historically agrarian political party and thus the party system as a whole.
This theory gives two empirical implications: 1) A rural/agrarian party which has developed before full democratization should fare electorally worse after democratization in areas with historical high levels of land inequality compared to areas with historical low levels of land inequality 2) The relative effect of historical land inequality on agrarian party vote share will be greater in areas where a larger share of the population lives in rural areas.
In the next section, I describe the case of early 20 th century Denmark which will serve as the empirical testing ground for this theory.
3 Rural parties and inequality in Denmark before and after democratization 3.1 Early 20 th century Denmark and its political parties To test the theory of historical land inequality and the electoral fortunes of agrarian parties after democratization, I use the case of Denmark in the early 20 th century, which at the time, like much of Northen and Western Europe, was still largely an agricultural society, where a large proportion of the population lived in rural areas. 14 During the formation of the modern Danish party system in the late 19 th and early 20 th century, the Liberal Party 15 emerged as the main agrarian party in Denmark, representing middle class farmers 16 and general rural interests in a country-wide organization with its own local newspapers, local party organizations, a nation-wide youth organization and strong ties to other rural and agricultural associations. While other parties, which emerged at the time, such as the Conservative Party and 14 Until around 1916, the majority (Fink 2000, 18). 15 In Danish "Venstre" literally "Left" an allusion to its opposition to the conservative regime at the time rather than a dedication to socialism. 16 Which, due to the export-orientation of Danish agriculture, were more classical liberal than rural interests in many other European countries.
the Radical Liberal Party 17 also had important rural constituencies, the Liberal Party emerged as the large broad-based rural party in Denmark in the decades before the Danish constitution of 1915 (Fink 2000, 14-19). 18 The constitution of 1915 introduced (almost) universal franchise for both men and women and marked Denmark's final transition to democracy. However, based on the theoretical arguments above, we might wonder what role land inequality played for the Liberal Party's build-up of local party organizations and partisan loyalty among the Danish rural population before democratization and, consequently, which role it played for the electoral fortunes of this party after democratization.

Land inequality in Denmark in the early 20 th century
While modern Denmark is often held up as a prime example of a society with a high level of socio-economic equality, rural economic life in Denmark during the 19 th and early 20 th century was characterized by extremely stark inequalities in the main economic asset of the time, agricultural land. Large landed estates, many owned by the Danish aristocracy, still dominated especially the Southern islands of Denmark, while a large share of rural households held little or no agricultural land and had to supplement their income with work on larger agricultural estates or by holding other part-time professions. 19 Based on a survey of agricultural land in 1905, the Gini coefficient for agricultural land in rural areas in Denmark was 0.76 giving Denmark in 1905 a higher level of land inequality than Germany in 1895 (Ziblatt 2009, 72). 20 However, there was substantial variation in land inequality among Danish regions (counties), confer figure 1. The Western part of Denmark had a much more equal distribution of agricultural land compared to the East and South of the country.
17 Which split from the Liberal Party in 1905. 18 The fourth main party which emerged during these decades, the Danish Social Democratic Party was in its early years mainly a party for urban workers.
19 Especially near the Danish capital of Copenhagen. 20 With the important caveat that the measurement of agricultural land was not identical for Denmark and Germany.

The impact of the land reforms
However, this very unequal distribution of agricultural land in Denmark did not continue long after Danish democratization. In the early 1920s, a number of land reforms dramatically distributed agricultural land away from the largest landowners to rural smallholders, which had previously held little or no agricultural land of their own.
Through the abolishment the so-called entailed estates 21 in 1919, which meant a confiscation of 20-25 percent of the total wealth and around 20,000 hectares of agricultural land from these estates, which were among the largest Danish landowners, as well as through a redistribution of church land 22 in Denmark, the Danish governments were able to construct thousands of smallholder farms during 1920s for people which had previously held little or no agricultural land (Erichsen and Tamm 2014). These reforms dramatically decreased land inequality in Denmark during this period.
These land reforms also seemed to have reduced land inequality to a much larger extent in areas previously characterized by very high levels of land inequality. In figure 2, I plot the agricultural land which was confiscated from the large entailed estates in the 1920s as a consequence of the 1919 land reform based on data from Erichsen and Tamm (2014) against the level of land inequality in 1905. There is a clear positive association between these two factors. The land reform of 1919, which redistributed over 20,000 hectares of agricultural land, hardly impacted areas with historical low levels of land inequality, whereas it had a much larger effect on the redistribution of land in areas with historical high and medium level of land inequality. Thus, these land reforms, apart from having greatly equalized land ownership in Denmark in general, also seem to have made differences with regards to level of land inequality smaller between Danish counties. Thus, by the 1920s and 1930s, the difference in land inequality between Danish counties seemed to have been reduced. However, as argued in the theoretical section, the historical differences in land inequality between Danish counties might still have had a substantial political impact, especially for the electoral fortunes of the main agrarian political party, in this case the Liberal Party. I now turn to the data to test this argument.

Data and estimation
The data used to test the effect of historical land inequality on the vote share of agrarian parties, consists of a panel of the main 18 Danish counties observed during all Danish lower-house elections from 1901 to 1943. During this period, Denmark introduced full franchise for both men and women in 1915 and proportional representation from 1920. 23 This was a dramatic change since the early 20 th century, where Denmark had relied on a first-past-thepost system with only voting rights for male head of households (Fink 2000, 69-75). However, as previous mentioned, all main political parties competing for votes during this period had their origins in the time of the limited franchise before 1915 (Fink 2000, 14-33). To measure the local level of support for the main rural-based party, I obtain data for the vote share of the Liberal Party (Venstre) in each Danish county from 1901 to 1943. The 1915 election was a so-called "peace election", where most candidates ran unopposed, so this election is excluded from the analysis. I later also look at the local vote share of other Danish political parties after democratization. The data for the vote share of the Danish parties during this period comes from Constituency-Level Elections Archive (CLEA) dataset (CLEA 2018) which is supplemented with data from the Danish Statistical Agency for the Danish Nazi Party in the 1935 election, where electoral data for this party is not available from the Constituency-Level Elections Archive.
The central independent variable is historical land inequality in each Danish county. In order to measure this, I calculate local Gini coefficients for agricultural land in Denmark using a unique survey of all agricultural land in rural districts 24 in Denmark in 1905 obtained from the Danish Statistical Agency's Statistical Yearbook of 1919. This survey records the number of farms in different size categories as well as the total agricultural land in each size category for each of the Danish counties at the time. Consequently, it is fairly easy to calculate a Gini coefficient for agricultural land for each of the Danish counties. 25 Since the four Danish counties of South Jutland/North 23 The capital region of Copenhagen had used proportional representation since 1915. 24 The calculation thus excludes the limited amount of agricultural land attached to Danish towns and cities. 25 The survey uses as its unit of classification an older Danish measurement of agricultural land called barrels of "hartkorn" (literally "hard grain") which takes into account not only the area but also the yield potential of the agricultural land.
Schleswig at the time of the survey were part of Imperial Germany, there is no land inequality data for these counties and these counties are thus excluded from the analysis. In appendix B, I redo the core analysis with an alternative measure of land inequality using the share of agricultural land in rural districts held by the farms/estates in the largest size category. 26 These results are largely similar to the main analysis.
As additional control variables in the analysis, I also obtain data for the rural population share in each county as well as the size of the county population, the latter which is included as a control variable in log form. Since part of the theoretical argument is that historical high land inequality decreases the later vote share of rural-based political parties relative to these parties' potential vote share in rural areas, in later estimations, I interact rural population share with the main difference-in-difference interaction. Data for both of these demographic variables are from different Statistical Yearbooks provided by the Danish Statistical Agency. Finally, based on data from Erichsen and Tamm (2014), I also code the number of hectares of agricultural land in each county which was redistributed away from the entailed estates after the 1919 land reform. Descriptive statistics for the variables of the main analysis can be seen in table 1. The estimation of the effect on Liberal Party vote share of historical land inequality after democratization is done by a difference-in-difference setup (Angrist and Pischke 2009, 227-243), where I interact historical land inequality with a dummy which takes the value 1 after democratization (the years after 1915) and 0 otherwise, while at the same time holding countyand year-effects constant. 27 The estimation itself is done by ordinary least squares with county-clustered standard errors. 28 The regression equation can be seen in equation 1 below. (1) Where V ote is the Liberal Party's vote share in county i in election year t, DL.Ineq is the difference-in-difference effect consisting of an interaction between the dummy D which takes the value 1 after democratization (after 1915) and historical land inequality L.Ineq. R.P op is the share of the county population living in rural areas. Log.P op is the log of the county population while δ i and γ t are the county-and election-fixed effects while it the error term. Since historical land inequality is time-invariant, it does not appear independent in the fixed-effects setup.
In some later estimations, to test the argument that the historical inequality effect should be larger in rural areas, the share of population in rural areas is added to the difference-in-difference interaction in order to explore whether the effect of historical land inequality after democratization is larger in rural areas. The regression equation then looks like this: 5 Results: Land inequality and Liberal Party vote share The results from the difference-in-difference estimation can be seen in table 2. In column 1, the simple difference-in-difference estimate without any controls can be seen. The results provides substantial evidence in favor of the 27 See also Andersen and Jensen (2019, 529) for an example of a similar type of differencein-difference set-up. 28 Since there are 18 counties, some issues might be raised about the relatively small number of clusters (Esarey and Menger 2019, 542-544). In appendix C, the core results are redone with un-clustered and bootstrapped standard errors. theory of land inequality, early party building and agrarian party electoral fortunes after democratization. While democratization seems to have a positive effect on the Liberal Party's vote share, this effect declines substantially in areas characterized by historical high levels of land inequality, as the interaction between democratization and historical land inequality is negative and statistically significant. In accordance with the theoretical argument, the Liberal Party's lack on an incentive and/or opportunity to build strong voter loyalty and strong party organizations in areas characterized by high levels of land inequality under a limited franchise had substantial implications for the electoral fortunes of this party after democratization. This effect does not change with the addition of controls for share of rural population and county population size in columns 2 and 3.
However, as previously argued, we should expect the electoral penalty of historical high levels of land inequality after democratization to be greater in rural areas, where the Liberal Party with its agrarian profile should have done better electorally but where high historical levels of land inequality might have decreased this rural electoral premium after democratization. In order to test this aspect of the theory, in column 4, the share of rural population variable is added to the difference-in-difference interaction. The results provides some evidence in favor of the theoretical argument, the three-variable interaction is negative, while the rural population and democratization interaction is positive, indicating that the Liberal Party did relatively better in rural areas following democratization but that this premium was smaller in areas characterized by high levels of historical land inequality. While the coefficients are not statistically significant at conventional levels, there is at least suggestive evidence for this aspect of the theory, confer also the marginal effects visualized in figure 3.
There is thus substantial evidence, in accordance with the theory of rural party building under a limited franchise, that the main rural party in Denmark, the Liberal Party, did worse electorally in areas characterized by historical high levels of land inequality after democratization. An effect which seems to be larger in rural areas. However, a concern might be that the difference-in-difference effect of historical land inequality after democratization instead is capturing other trends in areas with historical high land inequality after Danish democratization. As evident from figure 2, the Danish land reform which distributed substantial levels of agricultural land away from the Danish aristocracy in the 1920s mainly happened in areas with higher levels of historical land inequality. In order to address this concern, in column 5, I add another difference-in-difference estimation by interacting the amount of land in each county which was redistributed in the 1919 land reform with a dummy which takes the value 1 if the land reform of 1919 has come into effect (from 1921). However, the results from column 4 remains even with this interaction.  A visualization of the difference-in-difference estimate regarding the effect of historical land inequality after democratization contingent on rural population can be seen in figure 3, which plots the interaction from table 2, column 5. The same trend as in table 2 appears. Before full democratization, there is little moderating effect of land inequality on share of rural population's effect on Liberal Party vote share. However, after the transition to democracy, historical land inequality strongly moderates the effect of rural population share on Liberal Party Vote share. In areas with low historical land inequality, rural population share has a strong and statistically significant effect on Liberal Party vote share, while this effect is much weaker in areas characterized by historical high levels of land inequality. In accordance with theoretical argument, the Liberal Party's weaker incentive to build strong local party organizations and partisan loyalty in rural areas with high levels of land inequality during the time of the limited franchise had substantial political implications after the transition to democracy.

Land inequality and the electoral fortunes of other parties
Apparently, the Liberal Party, which developed in Denmark to represent rural interests before full democratization, did much worse electorally after democratization in areas where land inequality was historically high. This effect seems to have been particular strong in areas with a relative larger rural population. A finding which is in accordance with the theoretical argument about how high levels of land inequality, under a political regime with a limited property-based franchise, disincentivizes rural parties from building strong party organization and launch mobilization campaigns in these highinequality areas during the party's initial formation. A factor which can have wider electoral consequences, even after full democratization and even after substantial changes to land inequality. However, the question naturally arises for which political parties benefited from the Liberal Party's relative weakness in areas with historically high land inequality?
In order to explore this issue, in this section, I analyze the combined effect of historical land inequality and rural population share on the electoral fortunes of other political parties after democratization. Since some of these parties did not exist or had lower incentive to run in certain areas before the introduction of the full franchise, and thus democratization, I only analyze the democratic period from 1915 to 1943 (first election year is 1918) using random effects estimation. The results can be seen in table 3, where the first column reports the effect of the Liberal Party. The results show substantially the same story as the difference-in-difference analysis, the interaction between historical land inequality and share of rural population is negative, while the coefficient for the rural population share is positive. After Denmark's transition to democracy, the Liberal Party did better in rural areas but this rural premium was lower in areas characterized by high historical levels of land inequality. I now turn to the electoral fortunes of the other political parties First, I look at the effect of the land inequality-rural share interaction on the vote share of the Conservative Party and the Radical Liberal Party, the two other main parties which arose before full democratization to represent rural interests, the Conservatives the large estate holders and the centre/centre-left Radical Liberal Party the rural smallholders (Fink 2000, 14-36). The Conservative Party had officially been founded in 1915 but was the direct descendent of the older Conservative Party formed in the 1880s. 29 The Liberal Party's weakness in rural areas with historically higher levels of land inequality seem somewhat reflected in better electoral fortunes of the two other parties historically representing other rural interests. The interaction between land inequality and share of rural population is positive both for the Conservative Party and positive for the Radical Liberal Party but only statistically significant for the Radical Liberal Party. 30 In column 4 of table 3, the dependent variable is the vote share of the Danish Social Democratic Party, which had developed to represent urban workers during the latter half of the 19 th century (Fink 2000, 16-19) and had mainly been politically represented in the larger towns and cities. Unsurprisingly, the urban-based party did poorer in more rural areas but this rural electoral penalty for the Social Democrats was much lower in areas characterized by higher levels of historical land inequality as indicated by the negative coefficient for rural population share but the positive coefficient for the land inequality and rural population interaction. However, neither of these coefficients are statistically significant. 31 In columns 5 and 6, I look at the effect of land inequality on the vote share of the more extreme parties in the Danish political system during this period, both on the left and on the right. In column 5, the dependent variable is the vote share of the Danish Communist Party in the 1939 election. The effect of the interaction between historical land inequality and rural population share found for the Danish Social Democrats seems similar for the Communist Party. While this party seems to have fared electorally worse in areas with 29 Named "Højre", literally meaning "Right" in Danish. 30 However, a difference-in-difference analysis with the Radical Liberals' vote share as the dependent variable provides somewhat less evidence in favor of them capitalizing on the Liberal Party's lack of electoral support in high-inequality rural areas after democratization. Results are available upon request. 31 During this period, the Danish Social Democratic Party became the dominant party in Denmark, also saw an ideological reorientation (towards more centrism) of the party and an appeal to broader segments in Danish society than urban workers (Knudsen 2006, 175-177). Table 3 provides at least suggestive evidence that this strategy seemed to have yielded better results in areas where the Liberal Party, due to historical higher levels of land inequality, had developed weaker partisan loyalty among the electorate. a large rural population, this penalty seems to have been reduced in areas with historically high level of land inequality. 32 However, the interaction is not statistically significant at conventional levels, which is mainly a feature of the low number of observations, as there is only data for this party in the 1939 election.
Finally, in column 6, I look at the effect of historically high rural inequality on the vote share of the Danish Nazi party (officially the Danish National Socialists Workers' Party, DNSAP), an extreme right party founded in 1930 with inspiration from the German Nazi Party, NSDAP (Lauridsen 2002, 55-56). Here, the interaction between rural population share and historical land inequality has a positive and statistically significant effect on this party's vote share. Apparently, the Danish Nazi Party did better electorally in rural areas characterized by higher levels of historical land inequality. These results suggest that the effect of historical land inequality on the electoral misfortunes of the Liberal Party, and the subsequent better electoral results of the centre-and radical left, do not merely reflect an effect of higher levels of historical land inequality on more leftist preferences in these areas, since the far-right also seemed to have benefited from this. These findings echo the findings of Charnysh and Ziblatt (2018) and Ziblatt (2017) who also argue and find evidence that the extreme right seemed to have benefited from the electoral misfortunes, due to local party organizational weaknesses, of liberal and conservative parties in Weimar Germany.
Apart from decreasing the relative electoral fortunes of the main agrarian party in Denmark, historical high land inequality also seemed to have increased party-system polarization, through an increase in the vote share for both extreme left and extreme right parties, after Denmark's transition to democracy. 33 Random effects regression except in column 5. Dependent variable is party vote share. County-clustered standard errors in parentheses. p < 0.1, * * p < 0.05, * * * p < 0.01.
The above results provides substantial evidence in favour of the theoretical argument that agrarian parties are disincentivized from building strong local party organizations and fermenting partisan loyalty in rural areas characterized by high levels of land inequality under a limited franchise, which have longer term consequences for these parties' electoral fortunes after the extension of the franchise. The main Danish agrarian party, the Liberal Party seemed to have done worse electorally in rural areas characterized by the historical high levels of rural inequality compared to areas with lower levels of historical land inequality. The beneficiaries of the low partisan attachment of the Liberal Party in these areas seemed to have been the centre-left and the extreme left, which generally did electorally worse in rural areas but less so in the historically unequal areas, as well as the extreme right. The above results are fairly robust to removing the capital region of Copenhagen from the analysis, confer appendix A. 34 7 Exploring the mechanism: Land inequality and the Liberal Party's pre-democratic organizational strength The above results indeed seem to suggest, in accordance with the theory of land inequality and early party formation, that the Liberal Party performed more poorly in areas characterized by historical high levels of land inequality after but not before democratization. One of the central mechanisms for this should be the weaker ability/incentive of this agrarian party to establish strong local party organizations in areas of high levels of land inequality under a system of limited franchise.This, due to the path-dependency of such organizations, should have left them less able to mobilize/having previously mobilized the new voter groups after the transition to the full franchise. How-34 With the exception of support for the Danish Nazi Party, where the interaction effect seems to be driven mainly by the Copenhagen area, which was one the party's main strongholds outside of (excluded) counties of South Jutland/North-Schleswig (Lauridsen 2002, 559). These countries had not been part of Denmark during the formation of Danish party system and the rural population there thus also had a relatively weaker history of association with the Liberal Party. Furthermore, historical land inequality (with no interaction with rural population share) is still positively and statistically significantly associated with Danish Nazi Party vote share both with and without Copenhagen included. These results are available upon request. ever, so far this mechanism has been left untested. Consequently, in order to test one of the central mechanisms behind the above results, in this section, I investigate whether higher local land inequality was indeed associated with less local organizational capacity of the Liberal Party before democratization.
Unfortunately, comprehensive data for the number of local party organizations and membership of these in Denmark at the beginning of the 20 th century does not exist. However, there is another source of data which reasonably captures the depth and breadth of the local party organizations in the early 20 th century, the number of local party-affiliated newspapers. Partisan newspapers were an extremely important part of the Danish political parties' organizations and mobilization efforts at the local level in the late 19 th and early 20 th century (Søllinge and Thomsen 1989, 67-68). While all of these party-affiliated newspapers were not necessarily owned by the national and/or local party organization, party-affiliated newspaper held close ties, including financially, to their respective political parties both nationally and locally and were usually members of the party's national political press organizations (Søllinge and Thomsen 1989, 67-69).
Thus, to measure the density of the Liberal Party's organizational capacity at the local level before democratization, I code, based on data from Søllinge and Thomsen (1989), the total number of local newspapers in each county affiliated with the Liberal Party in 1906, 35 the year after the split between the Liberal Party and the Radical Liberal Party. 36 I then construct a variable measuring the number of Liberal Party-affiliated newspaper to the total population. I then regress the newspapers/population variable on the land inequality (1905) variable. These results can be seen in table 4.
The results in table 4 clearly show a negative statistically significant association between land inequality and the number of Liberal Party-affiliated newspaper to the county population in 1906, even when controlling for the share of rural population and the total number of inhabitants in each county. 35 In general, I only code a newspaper as affiliated with the Liberal Party if it is the official affiliation of the newspaper. Ideologically/interest closeness is not considered sufficient to be coded as affiliation. The exception is "Agrardagbladet" in two counties which is considered sufficiently closely associated with the Liberal Party to be coded as affiliated.
36 Which also meant an allegiance shift of some of the former Liberal Party-affiliated newspapers. In accordance with the theory of land inequality and rural party formation, high disparity in agricultural land at the local (rural) level would have disincentivized and made it harder for the agrarian Liberal Party to build strong local party organizations, including local party newspapers, during the critical juncture of its initial formation, which turned out to have substantial political implications after the transition to democracy.

Discussion and conclusion
How does the economic structure of a country, including the level of inequality, affects partisan loyalty and attachment among the electorate during the critical period of party system formation? This article has argued that when party system formation takes place under a limited franchise, political parties representing rural interest have lower incentive to build strong local party organization and cultivate voter attachment in areas characterized by high levels of land inequality. This later impacts their electoral fortunes after democratization. This theory is tested on a novel dataset of Danish counties in the decades before and after Denmark's transition to democracy. In accordance with the theory, the agrarian Liberal Party, which emerged to represent rural interest before Danish democratization, did much poorer electorally after democratization in areas with a historically high level of land inequality compared to areas with a historically lower levels of land inequality. Instead, rural areas with high levels of historically land inequality saw greater electoral support for both the moderate and far-left parties as well as for the extreme right, which suggest that a legacy of high land inequality might also produce a more polarized party system.
These results hold substantial implications for the study of political parties, electoral volatility and party systems in general both with regards to the historical determinants of party systems of Western and Northern Europe but also for democratizing countries today as well as for recent political developments in established democracies.
The results of this article strongly suggest that economic inequality affects domestic politics, including the nature of the party system, in conjecture with political institutions and that the nature of these factors during critical junctures such as party system formation, might have substantial political impact even decades after. While this article has focused on parties with a rural appeal and land inequality, the overall logic might also apply to other types of parties and other types of socio-economic inequality. Thus, the level of inequality and the structure of political institutions at the time of a country's transition to a multi-party political system, either during or before democratization might have substantial impact on the ability and incentive of new political parties to cultivate voter attachment and build formal party organization and thus substantial impact on the stability of the party system even long after the process of democratization is over. Future research on party systems in new democracies could pay more attention to these issues.
This article thus speaks to the wider discussion about the political impact of socio-economic inequality (Albertus and Menaldo 2016;Scheve and Stasavage 2017). While the issue of whether and how socio-economic inequality affect party system polarization is still the subject of some debate (Fenzl 2018), the results of this article suggest that historical levels of inequality might have a substantial impact of the current state and nature of the national party system, even after these inequalities have disappeared or lost their relevance.
Finally, the article's overall theoretical logic about how the (lack) of incentive for political parties to build strong local party organizations and partisan loyalty is a function of both political institutions and socio-economic factors, which might have substantial consequences when political institutions change, might be more broadly applicable. Thus, this article's theoretical perspective might be able to shed light on other dynamics of voter attachment to political parties beyond periods following democratization. For an example, can the rise of the UK Independence Party and its relative popularity among working class voters in areas historically dominated by the Conservative Party in the United Kingdom (Mellon and Evans 2016), be explained by the historical lack of an incentive for the British Labour Party to cultivate local blue-collar worker attachment and build strong local party organizations in these areas, due to the incentive provided by the UK singlemember district electoral system? A decision which later, when the issue of European Union politics became salient in UK politics and the use of proportional representation to the European Parliament changed the political opportunity structure, suddenly drastically transformed the British political landscape?
The perspective laid out in this article also suggests that present economic factors, including inequality, as well as very specific features of political institutions might determine to what extent new political parties in modern Western Europe, including left-wing and right-wing populist parties, will be able and/or incentivized to cultivate longer-term attachment and loyalty among their voters beyond their immediate electoral gains. 37 . As suggested by current research on populism and inequality, the link between inequality and the propensity to vote against established institutions and parties is subtle and has an important geographical component (Adler and Ansell 2019;Ansell et al. 2019). The findings of this article corroborate this view but add that the specificities of political institutions during critical junctures of party formation might matter a lot for how inequality affects later political outcomes. Furthermore, the findings also suggest that historical inequalities, in place during these critical junctures, might cast a long shadow even when these inequalities have diminished and/or have lost their economic and social relevance. Future research could address these issues.
37 See Voogd and Dassonneville (2019) for research on this issue. See also Han (2016) and Rooduijn and Burgoon (2018) for the role of inequality for populist party support.   Random effects regression except in column 5. Dependent variable is party vote share. County-clustered standard errors in parentheses. p < 0.1, * * p < 0.05, * * * p < 0.01.  C Core results with un-clustered and bootstrapped standard errors Dependent variable is Liberal Party vote share. Bootstrapped standard errors in parentheses. * p < 0.1, * * p < 0.05, * * * p < 0.01.