Unequal Geographic Representation in a Mixed-Member Electoral System: Evidence from the German Bundestag

Abstract

This paper demonstrates a systematic geographic bias in the German mixed-member electoral system. This bias concerns the composition of the individual party groups, and, by extension, the composition of the parliament: The Bundestag is much more urban than it would be under equal geographic representation. The bias is caused by the distribution of list seats across districts: since parties have incentives to give the best list positions to candidates from their strongest districts, regions where the vote is more fragmented systematically get more representatives. Since the vote is more fragmented in urban districts, these districts have more MPs than rural districts. The paper empirically analyses Bundestag elections between 1994 and 2017 and demonstrates that geographic disproportionality has increased strongly with the growing fragmentation of the German party system. This affects substantive representation in terms of leadership positions and voting behaviour in the Bundestag. Unequal representation in the German political system therefore also has a geographic component.

INTRODUCTION

A number of high-profile political events such as Brexit, the Gilets Jaunes, and several national elections have raised political scientists’ interest in patterns of conflict between urban and rural areas. Many studies have demonstrated increasing differences on the demand side of the political system, that is in political attitudes or political behaviour between urban and rural dwellers (e.g. Jennings and Stoker 2016; Huijsmans et al. 2020; Gimpel et al. 2020). These studies diagnose a conflict that is based on both an economic divergence between booming metropolitan areas and the left-behind countryside (Rodríguez-Pose 2017; Iversen and Soskice 2019) and a cultural clash between ‘cosmopolitan’ cities and ‘nationalist’ towns and rural areas (Maxwell 2019; Cramer 2016).

At the same time, there is growing interest in the supply side of the political system and in the political representation of urban and rural interests. Here, electoral systems play a crucial role. That’s on account of the translation of votes into seats being typically structured along geographic lines. In this context, the increased intensity of urban-rural conflict seems to hold bad news for urban parties and their voters, since there is wide agreement that electoral rules tend to disadvantage them. This is most clearly the case in majoritarian electoral systems. Since urban voters are distributed inefficiently in geographical space, an electoral system that divides mandates equally across geographic districts will put them at a disadvantage: even if urban parties win a majority of votes, they often fail to win a majority of seats (Rodden 2019). However, even in PR systems, the urban left is often disadvantaged if district size varies within countries, since urban voters tend to cluster in bigger and therefore more proportional districts (Kedar, Harsgor, and Sheinerman 2016; Monroe and Rose 2002). The best the urban left can hope for, it seems, is the geographic neutrality promised by single-district PR systems such as the Netherlands and Israel.

In this paper, however, I study an important exemption from this general rule. As I will demonstrate, mixed-member systems, such as the German system, actually tend to overrepresent urban electoral districts – and thus urban voters – in parliament. This overrepresentation does not concern the strength of different parties but the composition of the individual party groups, and, by extension, the composition of the entire parliament. While urban districts are systematically represented by a higher than proportional number of MPs, rural districts have a less than proportional number of MPs.

To measure the German Bundestag’s geographic disproportionality, I count the number of MPs that have run as candidates in each district. This is possible because almost all German MPs run as district candidates in the majoritarian tier of the German system.¹ Since there are twice as many seats as districts (disregarding overhang and compensation seats), districts should on average have two MPs.

Using this definition, Figure 1 shows that disproportionality has increased enormously over the last four decades. The share of districts with exactly two representatives has declined from almost 60 per cent in 1980 to just 37 per cent in 2017. At the same time, the number of districts with three or more representatives has more than doubled from 17 per cent to 38 per cent. In the last Bundestag, 11 per cent of districts had four or more representatives, a category that did not even exist in 1980. At the same time, the number of districts with just one MP also increased – although this trend was reversed in 2017 due to the massive increase in the total number of MPs that was caused by a record number of overhang and compensation seats.

Figure 1 NUMBER OF MPS PER DISTRICT, SELECTED YEARS

The number of MPs per district has, thus, become much more unequal in recent years. The main claim of this paper is that this dispersion does not occur randomly. Instead, there are systematic patterns leading to an overrepresentation of urban districts.

Why is this the case? In a nutshell, the paper argues that the German mixed-member system creates a premium on electoral fragmentation, since parties tend to nominate candidates from their regional strongholds to the top positions of their party lists. Therefore, regions in which several parties have a relative stronghold will be represented in the top list positions of several parties. In contrast, regions that overwhelmingly vote for a single party will only be present at the top of a single party list.² As a consequence, fragmented regions will be represented by more MPs than regions that vote uniformly. When the urban vote is more fragmented than the rural vote, as is the case in Germany, this leads to an overrepresentation of urban regions in parliament.

The overrepresentation discussed so far is a formal overrepresentation: urban districts have on average more MPs than rural districts. This formal overrepresentation, however, may also lead to a substantive overrepresentation. In this regard, the geographic disproportionality in mixed-member systems may be less consequential than the disproportionality in majoritarian systems since it usually does not affect government formation (which occurs on the level of parties). Nevertheless, it deserves attention. After all, a large part of parliamentary politics is below the level of government formation and law-making. Which issues make it to the agenda? Which party members are elected into leadership roles or get important committee assignments? Do parties build mutual trust with other parties with whom they are not in a coalition? Which experts are invited to testify in important hearings? These and a myriad of other questions are affected not just by the strength of different party groups, but also by their composition.

Disproportionality in numerical representation will have substantive effects when MPs consciously or unconsciously act as the representatives of certain geographical areas. They may simply do so because of strong personal ties to a locality: the place where they live, where they grew up, or where they are active in local politics (Folke et al. 2021; Gschwend and Zittel 2018; Zittel, Nyhuis, and Baumann 2019). In mixed-member systems, however, they also have explicit institutional ties to a specific geographic constituency: the district in which they ran as district candidates. Indeed, the empirical literature on Germany finds that MPs seek to represent the interests of their home districts, not only when elected as district candidates but also when elected as list candidates (Manow 2015; Siefken 2013, 2016). District MPs may be even more committed to their home districts (Stratmann and Baur 2002; Sieberer 2010), but what matters is that list MPs also have home districts which they seek to represent politically.

Whereas the existing literature focuses on the behaviour of individual MPs, however, this paper puts the district at the centre of analysis. My aim is to describe and explain which types of German districts are over- or underrepresented. I start by outlining what geographic representation means in proportional systems in general and in the German mixed-member system in particular. Based on this discussion, I then develop a theoretical argument about the importance of party strongholds in the German system and the resulting premium on electoral fragmentation. This argument allows me to derive specific hypotheses about which districts will be overrepresented. Finally, I review the different ways in which Bundestag members directly target their home-districts and hypothesise about measurable effects of geographic disproportionality on substantive representation.

The empirical part then proceeds in four steps. Firstly, it demonstrates that more fragmented districts systematically have more MPs. Secondly, it shows that this is the case because parties systematically nominate candidates from their strongest districts to the top of their lists. Thirdly, it shows that urban districts, particularly in university towns, tend to be overrepresented at the expense of districts in rural areas. Finally, it provides evidence that unequal geographic representation indeed leads to unequal substantive representation by analysing the geographic distribution of leadership roles and voting behaviour on moral issues. The conclusion discusses the potential implications of unequal geographic representation in the context of an emerging rural-urban divide in the German and many other political systems.

THEORY

For the purposes of my argument, it is helpful to conceptually distinguish between two forms of geographic representation. The first is formal geographic representation: how many MPs are formally representing a specific geographic unit? The second is substantive geographic representation: do MPs seek to represent the concerns and interest of specific geographic units?

In this section, I first explain my definition and measurement of formal representation in the German mixed-member electoral system. Afterwards, I develop an argument about which types of regions will be over- or underrepresented. Finally, I explain why unequal formal representation is likely to lead to unequal substantive representation.

Geographic Representation in the German System

Formal geographic representation is the concept of representation that is at the heart of majoritarian electoral systems, where each MP is the representative of a precisely defined geographical area: the electoral district. The formal representation of districts in turn is supposed to lead to their substantive representation: MPs are expected to represent the concerns and interests of the voters in their districts. However, this may lead to a substantive misrepresentation of certain national coalitions of interests when they are distributed inefficiently across districts. This usually generates a bias in favour of the rural right – and a well-known incentive to manipulate district borders. No matter how ingeniously districts are designed, however, each citizen is formally represented by exactly one local representative.

This formal type of geographic representation generally plays a much smaller role in proportional list systems which focus on the representation of national voter coalitions. After all, there is only a much weaker formal link between MPs and specific geographic constituencies. When an electoral system uses many, relatively small districts, MPs may still with some justification be seen as representing their district. This, however, becomes less plausible when district size increases.

Nevertheless, a growing number of studies have shown that a substantive form of geographical representation, in which MPs seek to further the interests of specific regions, also exists in proportional list systems. However, this form of substantive representation is typically based less on formal institutional ties and more on an informal form of ‘local representation’. Hence, research on this phenomenon typically studies legislators’ place of residence in order to measure substantive geographic representation (Fiva, Halse, and Smith 2019; De Maesschalck 2011; Folke et al. 2021; Latner and McGann 2005).

Mixed-member systems like the German system are particularly interesting in this context since they seek to reconcile the representation of districts with the representation of national voter coalitions. In German federal elections, each voter has two votes. With the first vote, the majoritarian element of the system, voters directly elect a candidate at the district level (there are currently 299 districts). These directly elected MPs are the formal representatives of specific districts and each district has one directly elected MP.

With the second vote, the PR element of the system, voters vote for a party at the level of federal states. Since district seats are deducted from the number of list seats for each party, the PR tier ensures the almost perfect proportionality of the system (Zittel 2018, 782; for a detailed treatment see Manow 2015). Conceptually, MPs elected through the party lists are representatives of national (or at least statewide) voter coalitions.

Empirically, however, the two logics of representation are not as strictly separated. As already described in the introduction, their interplay may lead to an unequal formal representation of different districts, whereby different districts effectively have a different number of MPs.

The reason is that more than 95 per cent of Bundestag members in recent elections have run either as pure district candidates or in both tiers of the electoral system. Such ‘dual candidates’ run as district candidates in a specific district and as list candidates in the district’s state. As a consequence, almost all MPs have a formal institutional link to a specific district. This formal link creates strong incentives to also represent the district substantively, e.g. by representing the policy positions of the constituents or by channelling resources into the district (see below).

In contrast to pure proportional systems, almost all German MPs are thus representing a precisely defined electoral district. In contrast to pure majoritarian systems, however, there can be more than one representative per district. Hence, formal geographic disproportionality can be measured by counting the number of MPs that ran in a district. This number ranges from a minimum of one (the district winner) to a maximum that equals the number of parties in the Bundestag if every party sends its district candidate as a list representative.

Compared to a pure PR system with a single, nationwide district, the German mixed-member system thus limits geographic disproportionality in two ways. The majoritarian tier of the system ensures that each district is represented by at least one member in the Bundestag. The proportional tier makes sure that all states are represented proportionally, since list seats are distributed by state lists. Within the confines of these proportionality-preserving elements, however, the distribution of list representatives can be geographically quite uneven. Imagine a small state like the Saarland, which consists of four districts and will thus usually have eight representatives in the Bundestag. In principle, it is possible that three districts of this state send one representative each, whereas one district sends five representatives to Berlin.

Which Regions get Overrepresented

The German electoral system can thus generate systematic patterns of formal geographic disproportionality. Yet, which types of regions are over- or underrepresented? I develop my answer to this question in two steps. Firstly, I argue that the institutional incentives for party behaviour create a premium on fragmentation, leading to an overrepresentation of areas with a more fragmented vote. Secondly, I argue that the cleavage structure underlying the party system determines where this fragmentation and, thus, overrepresentation occurs. Given the specific structure of the German party system, I expect an overrepresentation of urban districts.

Why Fragmented Regions get Overrepresented

The first part of my argument is institutional: In closed list systems, the crucial determinant of the representation of different regions is the formation of party lists. If parties distribute their candidates equally across space, all regions should on average have a proportional number of representatives. If there are geographic patterns in the list formation, however, they translate into similar patterns in the geographic make-up of parliaments.

There is enormous variation in the details of list formation between countries, in particular with regard to the distribution of power between the central party office and local party organisations (Hazan and Rahat 2010). In Germany, however, differences between parties are relatively minor, partly because the German party law makes relatively detailed prescriptions for nomination procedures (Höhne 2010), and all parties use a delegates- and region-based nomination procedure (Spies and Kaiser 2012). Independent of these institutional details, parties will likely tend to nominate candidates from their relative strongholds, i.e. their strongest districts to the best list positions. As Latner and McGann (2005) argue, this can happen for two reasons: a vote maximisation logic and an internal party competition logic. According to the first logic, parties will nominate candidates from their strongest districts, since this is where the most votes are at stake for them. This logic is arguably even more powerful in the German system, since there is a potential for so-called ‘contamination’ where a strong district candidate also helps the party vote in the district (Manow 2015). According to the second logic, parties will nominate candidates from strongholds because the local chapters in electoral strongholds typically have the largest membership and thus the biggest influence. Stronghold party chapters will thus push through their preferred candidates in the internal nomination struggle.

As the empirical literature on the formation of party lists in Germany shows (Ceyhan 2018; Steg 2016), there are several other logics at work as well. For example, parties also seek a balance in terms of gender, party wings, and regions. These logics could potentially counter the stronghold logic, if strongholds are clustered in certain regions or if certain types of candidates are more likely to emerge from stronger districts. This is thus ultimately an empirical question and I provide empirical evidence for the systematic relationship between vote shares and list positions below.

The fundamental theoretical implication of this preference for nominating candidates from strongholds is that parties’ nomination strategies create a premium on fragmentation and lead to an overrepresentation of fragmented regions. This is because fragmentation of big ideological currents into several smaller parties should lead to correlated strongholds. Imagine that a party splits up into two parties with a similar regional distribution of votes. What used to be the stronghold of one big party now becomes the stronghold of two smaller parties. If both parties still put candidates from their respective strongholds to their most attractive list positions, this creates a premium on fragmentation. Regions that would send one representative through the big party will now send two representatives through the smaller parties.³

In a pure PR system, this tendency may be counterbalanced by the nomination strategies of the other parties. A party that dominates a non-fragmented region may simply nominate several candidates from that region to top positions on the party list. Hence, the fragmentation premium is much less likely to occur in pure PR systems. In the German system, however, such counterbalancing is not possible since the system limits parties to running one candidate in each district. Even if a party wins two-thirds of the vote, it can still only win one seat. The clearest example of larger parties’ inability to act strategically is the CSU: In 2013 and 2017, it won all districts in Bavaria. Accordingly, it got one representative from each district, with no differentiation between strongholds and relatively weak areas.

Based on these considerations about party strategies, I hypothesise:

Hypothesis 1: More fragmented districts tend to be overrepresented in the Bundestag.

Why Urban Regions get Overrepresented

This institutional argument thus predicts that regions in which the vote is more fragmented will be overrepresented in the parliaments of mixed-member systems. However, it makes no prediction about which geographical areas benefit from this fragmentation. If the rural vote were more fragmented than the urban vote, this should lead to an overrepresentation of rural areas in parliament, and vice versa.

Where the fragmentation is most pronounced will depend on the number and structure of cleavages that structure the party system. It is thus not possible to derive a general law of regional fragmentation. Still, it seems plausible that fragmentation in most contemporary European party systems is higher in urban than in rural areas. After all, the greater social and economic differentiation in cities and the higher diversity of the urban population should also lead to more heterogeneous political preferences. Indeed, the main transformations of European party systems over the last one-and-a-half centuries have predominantly affected urban parties: first, the replacement of Liberal by Social Democratic parties as the main representatives of urban interests; later, the rise of green and new left parties at the cost of Social Democrats. Today, cities are often strongholds of Social Democratic, Liberal, Left, and Green parties at the same time. Rural areas, by contrast, are usually strongholds of just two party families: the mainstream right and the radical right. This is also the case in Germany, where the fragmentation of the party system predominantly occurred on the political left and, therefore, in urban areas.

Moreover, the candidates in urban districts may also be stronger candidates in terms of having greater political ambition, better networks or access to more resources.⁴ This argument is developed for the US by Gimpel, Lee, and Thorpe (2011), who argue that the greater number of political organisations, as well as the greater access to fundraising opportunities and media networks facilitate the emergence of stronger candidates in urban areas. While not all of these mechanisms may fully apply to Germany, they would provide another source of an urban bias in party nominations.

I thus hypothesise:

Hypothesis 2: Urban and university districts tend to be overrepresented in the Bundestag.

Effects on Substantive Representation

The Bundestag is thus potentially characterised by inequalities of formal geographic representation. Yet, do these inequalities matter? Does numerical overrepresentation lead to substantive overrepresentation?

The comparative literature on geographic representation has found that even in purely proportional systems representatives seek to represent their local constituency, both by acting in line with their policy preferences and by obtaining resources. Legislators in multimember districts do not target all areas of the district equally but focus their efforts on specific sub-constituencies (André and Depauw 2018; Crisp and Desposato 2004). Evidence of local representation can also be found in patterns of political speech (Fiva, Halse, and Smith 2019; Fernandes, Won, and Martins 2020). Moreover, geographic representation may have concrete effects on policies. For example, areas in which many Swedish local politicians reside issue more building permits for one-family houses but fewer permits for multiple-family houses (Folke et al. 2021).

If legislators seek to cater to specific geographic sub-constituencies in pure PR systems, such effects should be even more pronounced in the German system, given the direct institutional link between representatives and districts. Indeed, a large literature has investigated whether this link only affects the behaviour of directly elected MPs or is also relevant for list representatives. This is a crucially important question since the allocation of list representatives creates geographic disproportionality.

Existing studies have clearly established that MPs indeed seek to represent their districts. Representatives devote a high share of their time to district work, are often assigned to committees which allow them to further the interests of their districts (Stratmann and Baur 2002; Heinz 2010), and bring up topics of particular interest to their local constituencies in parliamentary questions (Zittel, Nyhuis, and Baumann 2019). In election campaigns, many district candidates seek to cultivate a personalised vote (Gschwend and Zittel 2015). These activities also seem to be rewarded by voters, who personalise their vote choices in response to personalised campaigns (ibid.) This arguably contributes to the existence of an incumbency effect, in which running a current member of the Bundestag as a district candidate also strengthens the party in the party vote (Manow 2015).

Importantly, though, the literature has had difficulties to establish clear differences in these behaviours between district MPs and list MPs. Some authors have found such differences, for example with regard to representatives’ committee assignment (Stratmann and Baur 2002) or their proclivity to deviate from the party line (Sieberer 2010). Other studies, however, have disputed the existence of a mandate divide. Manow (2013), for example, shows that list MPs also seem to select committees that are particularly relevant for the districts in which they are running. The most recent research on MP defections finds that only a small, very specific subset of district MPs is more likely to deviate from the party line (Sieberer and Ohmura 2019).

Moreover, both types of representatives take into account the preferences of their constituencies when voting on a moral issue (Baumann, Debus, and Müller 2015). Studies of MPs’ time use and their district activities have found no relevant differences between district and list MPs (Siefken 2013, 2016). In interviews with MPs, Coffé (2018:, 376) finds ‘not much difference in the propensity of list and district MPs mentioning a district as the primary group of representation’. Finally, Zittel, Nyhuis, and Baumann (2019) find that local ties (like growing up in a district or being politically active on the local level) are more important than the mode of election.

These findings suggest that list representatives with dual candidacies also care for their district. Indeed, they have strong incentives to do so, since local party organisations have an important role in the formation of party lists. Hence, MPs who neglect their nominal district may run the risk of losing their spot on the party list (Manow 2012).⁵ Moreover, it seems that citizens also recognise the local activities of list MPs, since small parties also benefit from running incumbent MPs in a district, even if these MPs are by definition list MPs (Manow 2015). Finally, the importance of district representation may not be the same across all districts but depend on their specific characteristics, such as the competitiveness of the district (Stoffel 2014) or the volatility of the district results (Zittel, Nyhuis, and Baumann 2019). Since electoral results have become much more volatile in recent decades, this would suggest that this form of representation is also becoming more important.

Together, these findings show that geographic representation is a meaningful form of substantive representation in the German political system. It probably plays an even stronger role for district MPs, but the evidence clearly suggests that it also matters for list MPs. Hence, from the perspective of a district, it is attractive to have more (list) representatives in Berlin. Indeed, the only existing empirical analysis that studies the effect of the number of MPs per district finds that districts with more MPs were more likely to be spared from the cutback of military facilities in the early 2010s (Zittel 2014). I thus hypothesise:

Hypothesis 3: The more representatives a district has, the better it is represented substantively.

EMPIRICAL ANALYSIS

I will now test these predictions empirically. As I demonstrated in the introduction, the variation in the number of MPs per district has increased enormously since the 1980s. I start the analysis by investigating how this development is caused by the increasing geographic fragmentation of the German party system.

Fragmentation and District Overrepresentation

To do so, Table 1 regresses the number of MPs per district on a measure of district fragmentation (the Effective Number of Parties (ENP) in the district).⁶ This measure varied from 3.0 in the district Cloppenburg – Vechta to 6.1 in the district Berlin Neukölln in the election of 2017 with a standard deviation of 0.54. Since the number of MPs per district ranges from 1 to 6, I use a simple OLS.⁷ In the regressions in Table 1, I pool the seven federal elections between 1994 and 2017 and include election fixed effects. Since seats are distributed by state, I cluster standard errors on the state level.

Table 1 RELATIONSHIP BETWEEN NUMBER OF MPS PER DISTRICT AND VOTE FRAGMENTATION

	MPs per district
	ENP		ENP relative to state
	(1)	(2)	(3)	(4)
	Fragmentation	Controls	Fragmentation	Controls
Fragmentation
Eff. No. of Parties	0.548***	0.448***
	(0.088)	(0.090)
ENP rel to state			0.702***	0.606***
			(0.115)	(0.111)
District importance
Turnout		1.971+		2.093**
		(1.088)		(0.664)
Elig. voters		0.002+		0.002+
		(0.001)		(0.001)
State capital		0.268*		0.183+
		(0.101)		(0.104)
State fixed effects	Yes	Yes	No	No
N	2151	2151	2151	2151
r2	0.077	0.093	0.082	0.098

All models contain election fixed effects. Standard errors clustered on state level in parentheses.

+ p<0.10 * p<0.05, ** p<0.01, *** p<0.001.

In a simple model with the fragmentation measure and state fixed effects only, fragmentation is highly significant. A one standard deviation increase in the district-level ENP is associated with about 0.25 more MPs from the district. In model 2, I add a number of measures for the political importance of a district: the number of eligible voters (measuring the size of the district), turnout (measuring the degree of political participation), and whether the district is in a state capital (measuring closeness to political power). While all three control variables are also associated with a higher number of MPs, the association with fragmentation remains highly significant, even if it becomes somewhat weaker.

In models 3 and 4, instead of including state fixed effects, I use the relative ENP, defined as the difference between district ENP and state level ENP, as main independent variable. This is pertinent, since seats are allocated on the state level, so fragmentation has to be considered relative to the rest of the state. As Figure 2, which is based on Model 4, illustrates, districts that are more fragmented than their state are indeed much better represented in the Bundestag.

Figure 2 RELATIONSHIP BETWEEN FRAGMENTATION AND PREDICTED NO OF MPS

Nomination Strategies and the Premium on Fragmentation

How does this overrepresentation of fragmented districts come about? In the theory section, I argued that it occurs because (small) parties systematically nominate candidates from their strongholds to the top of their party lists. To test whether this is indeed the case, I use a dataset on all Bundestag candidates of the six major parties in the four elections between 1998 and 2009, generously shared by Philip Manow.⁸

The relative value of a specific list position depends on the size of the list, i.e. the size of the state. Moreover, district performance has to be defined relative to the state mean, since seats are distributed by state. I thus standardise my dependent and my main independent variable by state to allow for a pooling of states.

Since parties may rationally run stronger candidates in their strongest districts, I control for two measures of candidate strength: incumbency, that is whether a candidate is currently a member of the Bundestag, and experience, measured by the number of prior elections in which the candidate ran for the Bundestag. I also control for the number of eligible voters in a district, since parties may give better list positions to candidates from bigger districts.⁹ Finally, I control for candidate gender and add election and state fixed effects.

Table 2 shows the results of simple OLS regressions for each party. As expected, the three smaller parties, which do not have realistic chances of winning district mandates (with the exception of the Left in some parts of Eastern Germany), systematically assign lower (i.e. better) list positions to candidates in their strongest districts. For the Greens, a district with a one standard deviation higher than average vote share can thus expect to receive an about 1/7th of a standard deviation improved list position. In a medium-sized state like Hesse with a list size in the low 20s, this corresponds to pretty much exactly one list position. As expected, incumbents and more experienced candidates also get stronger list positions, while district size does not seem to play a major role.

Table 2 DETERMINANTS OF CANDIDATES’ LIST POSITIONS

	Standardised list position
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	Greens	FDP	Left	CDU	CSU	SPD	SPD Bav
Green district result	−0.147**
	(0.027)
FDP district result		−0.100**
		(0.024)
Left district result			−0.168**
			(0.034)
CDU district result				−0.009
				(0.019)
CSU district result					−0.094+
					(0.049)
SPD district result						0.077**	−0.089*
						(0.020)	(0.039)
Incumbent	−0.824**	−0.798**	−0.585**	−0.434**	−0.105	−0.533**	−0.658**
	(0.091)	(0.093)	(0.115)	(0.046)	(0.161)	(0.051)	(0.115)
No of candidacies	−0.169**	−0.154**	−0.180**	−0.104**	−0.097**	−0.096**	−0.190**
	(0.028)	(0.020)	(0.028)	(0.010)	(0.028)	(0.012)	(0.028)
Male	0.288**	0.395**	0.397**	0.307**	0.298**	0.305**	0.122
	(0.057)	(0.059)	(0.077)	(0.042)	(0.095)	(0.040)	(0.075)
District size	−0.001	−0.002*	−0.001	−0.001	0.001	−0.001	0.001
	(0.001)	(0.001)	(0.002)	(0.001)	(0.002)	(0.001)	(0.001)
N	654	1028	463	852	74	1176	179
r2	0.466	0.441	0.353	0.527	0.424	0.413	0.666

All models contain election and state fixed effects. Standard errors in parentheses.

+ p<0.10, * p<0.05, ** p<0.01.

For the major parties CDU, CSU and SPD, the picture looks very different. In fact, for the SPD, stronger district results are associated with worse list positions. This is because these parties follow a very different logic when forming their lists: their strongest candidates are effectively guaranteed to win their district and do not need the list as a fallback option. In a context where these parties are unlikely to win many districts, such as Bavaria for the SPD, they also follow the smaller parties’ pattern (model 7).

The Overrepresentation of Urban Districts

I now seek to connect the general observation of a premium on fragmentation with more specific characteristics of the overrepresented districts. The main premise of my hypothesis was that the vote is more fragmented in urban districts than in rural districts. This is confirmed by Table 3, in which I regress the district ENP on three indicators that capture different elements of urbanisation for each election between 1994 and 2017. These indicators are the log of population density, the population share younger than 35, and the presence of a university (data from Apfeld (2019)). As the table shows, districts with higher population density and with a university are systematically more fragmented. Moreover, this effect seems to have increased since 2009. The share of the young population, by contrast, has no systematic effect, when the presence of a university is controlled for.

Table 3 RELATIONSHIP BETWEEN POPULATION DENSITY AND DISTRICT FRAGMENTATION

	Effective Number of parties per district
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	1994	1998	2002	2005	2009	2013	2017
Pop density (ln)	0.129***	0.077***	0.114***	0.109***	0.196***	0.243***	0.309***
	(0.018)	(0.020)	(0.019)	(0.019)	(0.020)	(0.018)	(0.020)
Population < 35	0.012*	0.021**	0.010	−0.006	−0.017*	−0.010	−0.001
	(0.005)	(0.008)	(0.007)	(0.007)	(0.008)	(0.007)	(0.008)
University	0.093*	0.078*	0.106**	0.120**	0.211***	0.240***	0.229***
	(0.038)	(0.039)	(0.038)	(0.040)	(0.046)	(0.044)	(0.052)
N	328	328	299	299	299	299	299
r2	0.636	0.811	0.806	0.809	0.690	0.801	0.816

State fixed effects in all models. Standard errors in parentheses.

+ p<0.10 * p<0.05, ** p<0.01, *** p<0.001.

Urban districts are thus indeed more fragmented. But are they also overrepresented? To test this, Table 4 regresses the number of representatives from a district on a number of district characteristics while adding election- and state-fixed-effects.¹⁰ I include three types of characteristics that may explain the over- or under-representation of a district.

1. The already used measures of urbanisation

2. The measures of the political importance of a district used in Table 1

3. Measures of the economic strength of a district: the unemployment rate and the number of regular jobs per thousand inhabitants

Table 4 DETERMINANTS OF DISTRICT OVER-REPRESENTATION

	MPs per district
	(1)	(2)	(3)
	Pooled	1994–2005	2009–2017
Urbanisation
Pop density (ln)	0.067+	0.013	0.104*
	(0.034)	(0.048)	(0.041)
Population < 35	0.012	0.007	0.024
	(0.015)	(0.016)	(0.038)
University	0.342**	0.351**	0.255
	(0.110)	(0.109)	(0.187)
District importance
Turnout	3.192*	2.133	4.878**
	(1.166)	(1.415)	(1.627)
Elig. voters	0.003**	0.002*	0.002
	(0.001)	(0.001)	(0.002)
State capital	0.001	0.010	−0.019
	(0.107)	(0.094)	(0.197)
Economy
Unemployment	−0.002	−0.004	0.034
	(0.010)	(0.014)	(0.022)
Regular Jobs	0.001**	0.001*	0.001+
	(0.000)	(0.000)	(0.001)
N	2138	1241	897
r2	0.127	0.099	0.162

All models contain state and election fixed effects. Standard errors clustered on state level in parentheses.

+ p<0.10 * p<0.05, ** p<0.01, *** p<0.001.

In Model 1, I pool all elections between 1994 and 2017. In this model, population density and a university are significantly associated with more representatives per district. Two of the three measures of a district’s political importance are also significant. Finally, there is some evidence of an overrepresentation of more prosperous districts.

In models 2 and 3, I split the dataset into the periods before and after the big increase in the association between urbanisation and fragmentation that was visible in Table 3. Model 2 pools the elections from 1994 and 2005. In this model, effects are much smaller and only the presence of a university remains significant. After fragmentation became more strongly associated with urbanisation starting in 2009, population density also becomes significant.¹¹

The analysis thus shows that two types of districts tend to be systematically overrepresented in the Bundestag. First, districts in which many people vote (a high number of eligible voters combined with high turnout). From a perspective of democratic representation, this is a welcome result, since it reduces the effective overrepresentation of voters from smaller districts. Second, however, urban districts – in particular those with a university – also tend to be massively overrepresented, especially since 2009.

Consequences: Substantive Overrepresentation of Urban Districts

As the theory section has demonstrated, there are good reasons to expect that a numerical overrepresentation of urban districts will also lead to their substantive overrepresentation. Empirically, however, these effects are difficult to show, since many likely effects, e.g. on agenda setting within party groups, are hard to observe.

The empirical literature on geographic representation focuses on three measures to capture at least some of these effects: parliamentary roles such as committee memberships or leadership positions (Stratmann and Baur 2002; Heinz 2010; Manow 2013), political speech (e.g. Fiva, Halse, and Smith 2019; Fernandes, Won, and Martins 2020; Zittel, Nyhuis, and Baumann 2019), and representatives’ voting behaviour (e.g. Baumann, Debus, and Müller 2015; Kauder and Potrafke 2019).¹²

Here, I briefly revisit two of those measures: leadership positions and voting behaviour. In particular, I seek to show two things. Firstly, districts with more MPs are also better represented in terms of leadership positions and thus arguably have greater influence on the parliamentary agenda. Secondly, this influence is indeed valuable since MPs tend to act in line with the preferences of their districts. Urban MPs behave systematically differently from rural MPs when given the chance to do so. This analysis is not meant to conclusively establish the effects of geographic overrepresentation. However, it suggests that the results presented above are of more than just theoretical interest.

Leadership Positions

I start with an analysis of parliamentary leadership roles, capturing the internal power dynamics within parliamentary groups. To some extent, it is simply a mechanical effect if districts with more MPs get more leadership positions and thus not very surprising. Still, it is important to show that a higher number of MPs does not just result in a higher number of backbenchers. After all, many of the theoretically most interesting benefits of overrepresentation, such as the potential impact on agenda setting, have to do with leadership and gatekeeping roles. While it is hard to observe these effects directly, leadership roles are an important mechanism through which unequal geographic representation may affect the parliamentary agenda. If parties consciously counterbalanced the skewed composition of their party groups when distributing leadership roles, the potential benefit of district overrepresentation would be much smaller.

Analysing the effect of unequal geographic representation on leadership positions requires studying positions that are open to all parties and not just to government parties (such as ministers or junior ministers). I thus study committee chairs and party-internal leadership roles. I define all members of a party group’s executive committee and all party-whips (parlamentarische Geschäftsführer) as belonging to the party leadership.

I analyse the two roles separately since the total number of committee chairs is largely fixed and these positions are distributed proportionally among parties in the Bundestag, whereas party-internal leadership roles are generated by each party so that there are more leadership roles when there are more parties in the Bundestag. I thus control for the total number of parties in the parliament when analysing the distribution of party leaders.

In Table 5, I analyse all legislative periods between 1994 and 2017. The dependent variable is the number of MPs with a leadership role from each district (ranging from 0 to 3 for both measures). I first show that districts with more representatives are also more likely to have representatives in a leadership role (models 1 and 2). This suggests that, from the perspective of an individual district, it is indeed advantageous to be represented by a higher number of MPs.

Table 5 DISTRIBUTION OF PARLIAMENTARY LEADERSHIP ROLES

	Role of number of MPs		Role of urban districts
	(1)	(2)	(3)	(4)
	Committee chair	Party leader	Committee chair	Party leader
No of MPs	0.078***	0.131***
	(0.010)	(0.011)
No of parties		0.033+		0.037
		(0.017)		(0.032)
Urbanisation
Pop density (ln)			0.012	0.004
			(0.014)	(0.016)
Population < 35			0.004	−0.001
			(0.004)	(0.004)
University			0.013	0.116***
			(0.028)	(0.032)
District importance
State capital			−0.091*	−0.021
			(0.038)	(0.043)
Turnout			−0.055	0.766*
			(0.328)	(0.370)
Elig. voters			−0.000	0.000
			(0.000)	(0.000)
Economy
Unemployment			−0.005	−0.006
			(0.004)	(0.005)
Regular Jobs			0.000	0.000
			(0.000)	(0.000)
N	2151	2151	2138	2138
r2	0.038	0.076	0.015	0.037

All models contain election and state fixed effects. Standard errors in parentheses.

+ p<0.10 * p<0.05, ** p<0.01, *** p<0.001.

In models 3 and 4, I analyse which types of districts are overrepresented in leadership roles. Here, more urban districts are somewhat overrepresented among committee chairs, but the effect is not significant. This may have to do with the fact that several committees are directly concerned with the interests of rural districts. Among party leaders, whose selection is less constrained by the need to represent substantive policy areas, university districts are clearly overrepresented.

Voting Behaviour

While leadership roles confer power, it is not guaranteed that leaders use this power to further the interests of their district. The numerical overrepresentation in parliament in general, and in leadership roles in particular, is only valuable if MPs seek to represent the substantive concerns of their districts. Finding direct evidence of such district-related behaviour is not easy. After all, few political decisions target specific districts. A rare example is Zittel (2014), who studies the change in military personnel stationed in each district due to a Bundeswehr reform decided in 2011. Zittel’s unit of analysis is the individual MP, not the district. However, he finds that MPs who shared their district with a higher number of other MPs managed to limit the personnel cutbacks in their district. Zittel interprets this as an effect of more intense political competition within the district. In the context of this paper, however, it would also seem to signal the increased political clout of a district with multiple MPs.

Given the rarity of such measures of direct resource allocation to a district, a popular measure of district-related activities of MPs is their voting behaviour. Here, I use two examples of voting behaviour to study whether MP’s behaviour is in tune with the preferences of their districts.

Studying voting behaviour is relatively difficult in the German context, since party discipline in the Bundestag is extremely high. Studies typically rely on so-called ‘conscience issues’ in which party discipline is explicitly suspended. Two prominent examples of such ‘conscience issues’ were the vote on the liberalisation of preimplantation genetic diagnosis (PGD) in 2011 (Baumann, Debus, and Müller 2015) and the vote on same-sex marriage in 2017 (Kauder and Potrafke 2019). On both policies, there is a likely split in public opinion between more socially conservative rural areas and more liberal cities.

I analyse both votes in Figures 3 and 4.¹³ While all parties were split on the PGD, only the Christian Democrats were split on same-sex marriage. Figure 3 analyses the vote on the PGD on July 7th 2011 in which a majority of 326 MPs supported a restrictive liberalisation of the PGD, with 260 MPs opposed. All parties were split on the issue, with a majority of the Social Democrats, the Left, and the Liberals supporting the bill, while majorities of the Christian Democrats and the Greens opposed it.

Figure 3 LOGGED POPULATION DENSITY AND SUPPORT FOR PGD

Figure 4 LOGGED POPULATION DENSITY AND SUPPORT FOR SAME-SEX MARRIAGE AMONG CHRISTIAN DEMOCRATS

I regress an MP’s voting decision on the population density of his or her district, partisanship, and the MP’s age, gender and mode of election. Figure 3 shows the estimated relationship between district density and predicted voting behaviour. By this measure, MPs from the most urban districts were about 30 percentage points more likely to support the reform than MPs from the most rural districts. 60 per cent of the 326 representatives from districts with a higher-than-median population density supported the bill. Among the 260 MPs from districts with a lower-than-median population density, by contrast, just 132 (51 per cent) supported the bill. If districts had been represented more equally, the vote would have been much closer.

Figure 4 repeats the same analysis for the vote on same-sex marriage. Here, the Social Democrats, the Greens and the Left unanimously supported the policy. Moreover, a quarter of Christian Democrats (71) supported the policy while three quarters (221) opposed it. Following Kauder and Potrafke (2019), I thus only analyse the behaviour of Christian Democrats.¹⁴ Again, I control for the MP’s age, gender and mode of election. As the Figure shows, there was again a strong difference in the voting behaviour of urban and rural MPs. Whereas the predicted probability to support the bill was less than 10 per cent for the Christian Democrats from the most rural districts, it was more than 50 per cent for the Christian Democrats from the most rural districts.

Both analysed examples concern voting behaviour on moral issues. This is the case because conscience issues on which party discipline is suspended typically are moral issues. On other issues, such as economic questions, parties typically agree on a common position and vote (almost completely) uniformly. This does not mean, that MPs’ geographical backgrounds do not affect their own or their parties’ positions. However, it means that this effect is much harder to study.

Nevertheless, it is quite likely that urban districts also benefit from their overrepresentation in other policy fields. However, it is possible that the effect on economic issues is stronger with regard to salience than with regard to position. For example, a disproportionally urban Bundestag should pay much more attention to question of rent and housing than a disproportionally rural parliament. However, this does not necessarily mean that all urban MPs tend to have the same position, since this issue is also highly contested within cities.

Taken together, these simple analyses suggest that the numeric overrepresentation of urban districts may also lead to a substantive overrepresentation of these districts. Urban voters have more direct representation in important parliamentary leadership roles. Similarly, on specific policy decisions where the preferences of urban and rural voters are likely to diverge, urban voters have better chances to get their preferences represented.

CONCLUSION

This paper has empirically demonstrated that the representation of different German regions in the Bundestag has become more unequal between 1994 and 2017. More fragmented districts tend to be overrepresented since they contain strongholds of several parties and stronghold candidates are favoured on party lists. Empirically, this fragmentation premium leads to both a numerical and a substantive overrepresentation of urban and university districts at the expense of rural districts.

In this conclusion, I want to highlight three potential implications of these findings. The first implication concerns the institutional consequences of different electoral systems. As the paper shows, even electoral systems that are ostensibly geographically neutral can generate substantial geographical biases. In fact, when geographic proportionality is analysed from the perspective of geographic units, rather than of parties, a proportional system can be even more disproportional than majoritarian systems. Mixed member systems, which are often thought to combine the best of two worlds (Shugart and Wattenberg 2001), may actually combine the worst of both worlds in this context: The proportional tier of the system generates enormous potential for geographical disproportionality, while the majoritarian tier generates institutional incentives for MPs to act as representatives of their geographic constituencies. Mixed-member systems may thus lead to parliaments in which geographic disproportionality leads to a bias in substantive representation. This leads to the second implication, which relates to the debate about an emerging rural-urban cleavage in many developed democracies. Germany is typically seen as less susceptible to this cleavage due to its strong federalism and a polycentric, manufacturing-oriented economy that prevents the development of deep regional economic divides. While such an argument suggests a less intense polarisation in the preferences of voters, my analysis emphasises that the supply side of the political system may also play an important role in this conflict. If parties rationally react to the increasing fragmentation of the party system by focusing on the nomination of candidates from their strongholds, the party system may increasingly consist of distinctly urban and distinctly rural parties, even if these parties may programmatically still seek to address voters across the whole country. The interplay of voting behaviour and electoral rules may thus create an urban-rural cleavage on the supply side, in particular since similar patterns are likely to emerge on the state level (see Däubler 2017 for an analysis of the state of Baden-Württemberg).

The third implication concerns the link between unequal geographic representation and unequal responsiveness. The urban bias on the supply side may not just be a bias in representation but may also distort responsiveness. Indeed, the familiar result that the members of the German Bundestag are not very representative of the German population (still typically male, without a migration background, and much wealthier and more educated than the general population) has often been linked to the finding that the Bundestag is much more responsive to the preferences of upper classes than to the preferences of lower classes (Elsässer 2018; Elsässer, Hense, and Schäfer 2018).

So, while the question of unequal representation in the Bundestag has received a lot of attention, the literature typically focuses on individual attributes of Bundestag members – their gender, their education, or whether they have a migration background. Much less attention is being paid to the geographic dimension of representation, that is where representatives come from and which districts they represent. My results suggest that studying geographic patterns of responsiveness may be a fruitful avenue for future research. After all, my findings clearly suggest that districts with more representatives will also be substantively better represented. If the interests of urban and rural districts increasingly diverge, this may thus become an important source of unequal responsiveness in the German political system.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.

Notes on contributors

Lukas Haffert

Lukas Haffert is a senior researcher at the chair for comparative political economy at the University of Zurich. His research focuses on questions of public finance, historical persistence, and urban-rural divides.

Notes

1 In recent elections, more than 95% of Bundestag members ran as district candidates.

2 Or of none, since safe district winners do not need the list as a fallback option. In its strongest form, this argument thus only holds for parties that do not hold safe district seats (see below).

3 Importantly, these are relative strongholds (compared to the party’s state-level performance) and not absolute strongholds, since the allocation of seats happens on the level of states.

4 I would like to thank an anonymous reviewer for drawing my attention to this mechanism.

5 Parties often assign MPs additional districts as so-called areas of service (‘‘Betreuungsbereiche’’). However, since they will only run in their ‘real’ district in the next election, MPs have strong incentives to pay more attention to their ‘real’ districts, Siefken (2016).

6 To calculate the ENP, I use the combined vote share of all parties not represented in the Bundestag per district.

7 As shown in the appendix, results are very similar, but harder to interpret, when I use an ordered logit.

8 For a detailed description of the dataset, see Manow and Flemming (2012).

9 Districts are allowed to deviate up to 25% from the standard district size before they have to be redrawn.

10 In the appendix, I show again that the results also hold when using an ordered logit. Moreover, I show that these results are primarily driven by strongholds of the Greens.

11 Again, results are very similar when I use an ordered logit instead of OLS.

12 Zittel (2014), studies a specific policy outcome: the retrenchment of German military facilities.

13 Based on logit models. Full regression results are reported in the appendix.

14 This analysis extends Kauder and Potrafke (2019) since they only study the behavior of the 223 directly elected conservative MPs. My analysis also includes list MPs that ran in a district.