EXPLAINING GERMANY’S ELECTORAL GEOGRAPHY
Evidence from the Eastern States
StevenWuhs and Eric McLaughlin
Political Science, University of Redlands
ABSTRACT: Partisan attachments and voting behavior in Germany today are
more volatile than in the past. This article tests the enduring influence of
social cleavages on voting relative to two other factors that account for party
performance: path dependent forces and spatial dependence. Drawing on original
data from the eastern German states, we explain support for Germany’s
main parties in the 2017 federal election. We find relatively weak evidence for
continued influence of social divisions for the major parties, but that support
for the radical right Alternative for Germany (AfD) did reflect underlying
cleavage structures. Additionally, we identify reliable effects of the historical
immigrant population on contemporary voting. We also see weak evidence of
lock-in political effects associated with German reunification, limited only to
the CDU. Most interestingly, we observe powerful and robust effects of spatial
dependence for three of the four parties we examine. We conclude that the
effects presented here should signal to scholars of parties and electoral politics
the need to incorporate history and geography into their analytical frameworks
alongside more traditional approaches, since eastern Germany may in
fact be less spatialized than western Germany or other country cases because
of the homogenizing efforts of the SED regime.
KEYWORDS: 2017 Bundestag election, cleavage structures, electoral geography,
path dependence, spatial analysis
Introduction
Conventional political wisdom holds that divisions within society are the
primary drivers of voting behavior in democratic settings like Germany.
Those deep-seated social cleavages are outgrowths of particular historical
conflicts that configured and reconfigured the political landscape—divisions
between Catholics and Protestants, for example, or ones that resulted from
the rise of urban elites.1
While many of those conflicts predated the onset of
democratic competition, social cleavages in Western Europe were later
articulated in the party system (e.g., the development of Protestant and
German Politics and Society, Issue 130 Vol. 37, No. 1 (Spring 2019): 1–23
© Georgetown University and Berghahn Books
doi:10.3167/gps.2018.370101 • ISSN 1045-0300 (Print) • ISSN 1558-5441 (Online)
Catholic parties or bourgeois parties), and for most of the twentieth century
Western European party systems were structured according to these underlying
social cleavages.2
How those cleavage structures interacted with one
another was likely to produce partisan winners and losers during that period.
Partisan attachments and voting behavior in Europe today are more
volatile as a result of the weakening influence of social cleavages on the
vote, giving rise to new party challengers on the libertarian left and the radical
right and weakening the attachment of voters to the parties that dominated
postwar politics.3
That said, while weaker than in the past, social class
remains a reliable predictor of the vote especially in places where it was
once powerful.4
In Germany, for example, the Social Democratic Party
(SPD) is not the stalwart advocate for trade unionists that it once was, but
manual workers remain more loyal supporters of the party than they are of
its competitors.5
Likewise, religious Germans remain the most reliable supporters
of the Christian Democratic Union (CDU).6
That is, while major divisions
within a country’s voting population may not structure the vote as it
once did, social cleavages do retain important explanatory power.
Cleavages are not alone in structuring the vote, however. Recent scholarship
points to two additional factors that may also account for party performance
on election day. First, critical junctures and path-dependent forces
may shape parties’ electoral fortunes. By this logic, parties and party systems
develop through sequences of key decisions that establish particular developmental
trajectories for the institutions.7
Those paths can, in turn, position parties
differently vis-à-vis the electorate, meaning that path dependent forces
can shape how parties perform. For example, the choices made by reformers
within Eastern European communist successor parties during the initial transition
period powerfully shaped their parties’ prospects in elections through
the 1990s.8
As parties in that region struggled, those that had developed more
robust organizations and amassed greater resources in the past also proved
more resilient.9
That is, both elite choices and organizational resources contributed
to political legacies that affected party performance.10
In addition to cleavages and broader historical forces, variation in party
support may be accounted for by geography and spatial factors. Spatial
dependence is a regular feature in any election where vote totals are contained
within areal units.11
It is hardly uncommon for parties in democracies
to perform better in some distinct regions and less well in others. Yet, spatial
dependence is too rarely taken seriously by scholars of parties and electoral
competition. One of this study’s aims is to illustrate the utility of adopting
modeling strategies that account for this dependence. Spatial dependence in
elections can come from one of two (not mutually exclusive) sources.
••• 2 •••
Steven Wuhs and Eric McLaughlin
First, neighboring districts may exhibit similar voting patterns because
they are exposed to some explanatory variable that does not align with district
boundaries. Spatial dependence occurs due to “common exposure”12
or
“place-based”13
effects. In the eastern German context considered here, these
could include key economic or political developments, such as the interwar
industrial corridors stretching through Saxony and Saxony-Anhalt often referenced
by German politicians and political analysts.14
Place-based effects
might also be associated with the distribution of ethnic German minorities
with particular political identities—interviews oftentimes turned to the persistence
of communities of Sorbians in some Saxon regions, Slavic pockets in
Brandenburg, Sudeten German expelled after 1945 from what is now the
Czech Republic, and East Prussians in Mecklenburg-West Pomerania.
Secondly, spatial dependence may result from processes of diffusion. In
this case, neighboring districts in an election might express similar levels of
support for a party because voters interact more frequently with those from
nearby districts or because a campaign rally held in one district influences
voters drawn from its neighbors. One might expect these sorts of effects to
result from migration patterns as well—either within states or between them.
While significant political attention has been paid recently to Muslim
migration into Germany, the eastern region has also featured significant
outmigration toward the west as well as urbanization and depopulation of
rural areas—perhaps as much as one-quarter of the youth in some areas.15
While it is not possible to be certain about which process is most responsible
for any observed spatial dependence, spatial econometrics offers tools
that help determine which process is most likely responsible and to account
for that process in regression modeling. Recent work in political science
exhibits a new awareness of the need to integrate spatial analysis into the
study of electoral competition.16
For that reason, below we use spatial econometric
techniques in addition to Ordinary Least Squares (OLS) analysis to
test hypotheses.
Postreunification East Germany
In this article, we examine the power of cleavage structures, critical junctures,
and spatial effects on voting in Germany, specifically the regions of the
former German Democratic Republic (GDR). It makes analytical sense to
ask these questions there because despite being embedded within a larger
political system, eastern Germany is a coherent political unit featuring a distinct
subnational party system today, as it has since reunification. As in the
••• 3 •••
Explaining Germany’s Electoral Geography
western regions of the country, the CDU and SPD are the dominant parties,
but the liberal FDP and the Greens are largely absent in the east while the
successor party of the East German Socialist Unity Party (SED), now called
Die Linke (The Left Party), is the region’s third political force. In the 2017
federal elections, the radical right Alternative for Germany (AfD) emerged
as a particularly strong contender in the east as well.
In addition, social cleavages across eastern Germany are quite similarly
subdued. The East German SED regime sought to weaken organized religion
as a rival ideology to Marxism-Leninism, and it was largely successful.
Levels of religious affiliation across the region today are quite low:
Thuringia in 2011 had the highest overall level (31 percent total, with 24
percent Protestant and 7 percent Catholic) and Brandenburg has the lowest
(20 percent total, 17 percent Protestant and 3 percent Catholic).17
By
comparison, significant portions of Bavaria, Baden-Württemberg and
regions surrounding Cologne have Catholic memberships upwards of 70
percent and central Germany’s Protestant corridor features membership
rates of more than 40 percent, according to the 2011 census.18
Class divisions
were also muted by the regime, which glorified the proletariat while
seizing assets of propertied classes and redistributing them. State-level
unionization data collected by the German national labor federation
(Deutsche Gewerkschaftsbund, or DGB) suggest only a loose pairing of
union membership and state-level SPD support in the east, and in fact there
was evidence of an unexpected association after reunification between
labor organization and CDU vote share (i.e., more industrialized areas featuring
higher levels of CDU, rather than SPD, support). In MecklenburgWest
Pomerania and Saxony-Anhalt, there does not appear to be an
association between union membership and either party’s fortune, suggesting
a complex interplay of cleavages and other variables.19
Public opinion researchers consider the eastern German electorate distrustful
of political institutions and of parties,20
something quite common in
postsocialist contexts21
and recent democratizers. Furthermore, its voters are
very weakly partisan, with more than two-thirds self-identifying as weakly
partisan or non-partisan.22
Taken together, these attributes of the eastern
German electorate raise important questions about what predicts actual voting
behavior across the region. Are constituency-level attributes like the
structure of the economy predominant? Are voters locked into particular
patterns of partisan support because of historical experiences? Or do underlying
spatial phenomena account for their choices on election day?
••• 4 •••
Steven Wuhs and Eric McLaughlin
Model Specification
To test the relationships among cleavages, legacies, spatial factors and the
vote, we built an original dataset based on census records, government
data, and data collected from political parties themselves. Empirically, we
analyze the 2017 election performance of the Christian Democratic Union,
its primary competitor the Social Democratic Party, as well as the performance
of the Left Party and the emergent AfD. Our observations in this
analysis are at the forty-nine federal electoral districts (Wahlkreis) carved
from the territory of the GDR (Brandenburg, Mecklenburg-West Pomerania,
Saxony, Saxony-Anhalt, and Thuringia). While the German census agency
collects county (Kreis) level data on most of the indicators that we consider
here, the occasional redistricting of county lines made analysis at that level
impossible without losing entire states in our sample.23
Social Cleavages
Given the socialist regime’s efforts to dampen divisions within the eastern
German population, we expect relatively modest effects of contemporary
class and religious cleavages across the east. That said, some areas within
the east demonstrate high levels of religious or labor mobilization (as in the
Catholic Eichsfeld region of Thuringia). Cleavages there, as elsewhere in
Germany, may still structure the vote.
Measuring social cleavages is a complex endeavor.24
In this case, we use
data aggregated at the level of the electoral districts (Wahlkreise) used in the
2017 federal elections.25
To tap into the class cleavage, we draw on the proportion
of the workforce covered through the social insurance programs for
agricultural and industrial workers, respectively, in the German welfare
state. As an operationalization of the class cleavage, this is imperfect because
it does not capture organizational dynamics like bargaining power that
union density implies. It does, however, capture the overall economic structure
of a locality, and in a context like eastern Germany where union density
rates are low overall, those data themselves may misrepresent the
nature of the class cleavage. Operationalized as such, we expect that districts
with higher levels of agricultural employment will have higher levels of support
for the CDU and should be less likely to support the SPD. For industrial
employment, we expect the opposite relationship. In our examination of
Left Party and AfD support, we do not have clear ex ante hypotheses. As an
alternative operationalization in earlier models, we considered the GDP contributions
of agriculture and industry. Both were highly correlated with
insurance data and did not produce significantly different results in the OLS
••• 5 •••
Explaining Germany’s Electoral Geography
models below.26
To approximate the religious cleavage, we rely upon the
proportion of a district’s population that identifies as neither Catholic nor
Protestant. While traditional analyses of that cleavage were premised on the
cleavage between Catholics and Protestants, German (and especially eastern
German) politics today are more marked by a secular-religious conflict,
making this operationalization sensible. In addition to these three structural
variables, and in recognition of the fact that they are limited in their ability
to capture various elements of cleavage structure, we also control for unemployment
levels. The addition of further variables to the model is, unfortunately,
not possible given the small number of cases.
We also incorporate a variable that measures the proportion of the district
population that is from what the German government considers a
“migration background.” We add this variable because of the presence of
the refugee crisis in the media and campaigns heading into the 2017 contest
and the importance of Chancellor (and CDU leader) Angela Merkel’s openness
to refugee resettlement. The variable itself measures immigrant and
nonimmigrant foreigners who moved to Germany after 1949 and those
with at least one parent who immigrated after 1949. As such, we acknowledge
that this variable does not capture the impacts of the recent migration
patterns that were central to the 2017 contest; electoral district data on those
migrant flows are not available. We include the variable nonetheless
because it may capture how xenophobia and cultural difference affect party
support, if those effects are in fact related to exposure (i.e., in communities
with higher numbers of migrants, we might expect higher AfD support).
Alternatively, in German communities with longstanding migrant populations,
contact may in fact decrease feelings of difference, and thus drive the
vote away from the AfD. For the other parties, the direction of potential
effects is less clear.
Legacies
The upheaval that eastern Germany experienced in 1989-1990 was the sort
of transformative moment that could produce enduring alignments of voters
and parties. For that reason, we hypothesize that local-level political constellations
from the transition period shaped how electoral politics unfolded
afterward. In some areas, the SED regime had a strong infrastructural presence,
while others operated with relative independence from the regime,
leading to the emergence of a particular geography of protest across the
region prior to reunification. This activism, concentrated in Berlin and the
state of Saxony, was notably absent in Brandenburg, which interview participants
suggested was a function of the density of party functionaries living
••• 6 •••
Steven Wuhs and Eric McLaughlin
just outside Berlin. Where it was stronger, activism shaped the formation of
party organizations in the reunification period, in turn, driving election
results not just through the 1990s but today as well.27
Given these legacies,
we control for turnout in the models presented below.
In order to study the effects of those transition-era forces on local politics,
we examine the bloc CDU, an opposition party the SED regime sanctioned
and tightly monitored during its decades in power.28
Though the
bloc CDU did not offer the possibility of contesting power in the GDR system,
it did offer a reprieve to those opposed to the regime—an organizational
space to which they could retreat. In a very few regions, the bloc CDU
was even allowed to govern locally. Regions (in this case, electoral districts)
with higher densities of bloc CDU members were more independent of the
regime, and more likely to engage in protest when movements began prior
to 1989.29
We expect that areas with higher concentrations of bloc CDU
members in 1990 will have higher, more resilient levels of CDU support
today. To capture that effect, we use local membership in the bloc CDU in
1990, on the eve of the only democratic election of the GDR Parliament. We
draw on primary documents collected at the Konrad Adenauer Foundation
archive to calculate federal electoral district-level CDU membership in 1990.
The GDR government carefully monitored the activities and memberships
of local bloc CDU organizations, leaving behind documentary evidence in
the form of maps featuring membership data. We overlaid those (Stadtkreise
and Landkreise) data on a contemporary electoral district map to “scale up”
those membership data.
Spatial Effects
The data generating processes that determine the outcomes social scientists
study do not always respect territorial boundaries, and so spatial dependence
is likely to be a feature in the analysis of any electoral data based on
areal units. With georeferenced data, testing for spatial dependence and
building better fit models that account for that dependence is straightforward.
In the ensuing analyses, we model elections in the former East Germany
using, where appropriate, models with spatial error terms (“spatial
error models”) and models with spatially lagged dependent variable included
as a regressor (“spatial lag models”). Doing so will produce models that
better explain outcomes and also allow us to call attention to where those
models are most and least effective.
••• 7 •••
Explaining Germany’s Electoral Geography
Methods and Results
In each of the four models we estimate below, we begin with a straightforward
ordinary least squares (OLS) model using the forty-nine federal electoral
districts of the former East Germany in federal elections. The dependent
variable for each model is the level of electoral support received by the party
in question. Due to Germany’s mixed voting system, we use party support
in the second of the two votes cast (the Zweitstimmen or second PR list vote).
As predictor variables, we include the independent variables discussed earlier:
agricultural and industrial insurance participation, percentage of the
district neither Catholic nor Protestant, the percentage of the district from a
migration background, bloc CDU membership from 1990, unemployment,
and voter turnout. Along with each OLS model, we report a full range of
spatial diagnostic measures with each OLS model, following Luc Anselin
and Sergio J. Rey.30
We report Moran’s I, a measure of spatial dependence,
for each regression’s residuals. Statistically significant Moran’s I’s suggest
that there is unexplained spatial dependence in the model; unsurprisingly,
all of the models we estimate below exhibit spatial dependence.
Following Anselin and Rey, we use Lagrange Multiplier (LM) diagnostics
determine whether a spatial respecification of the OLS model is appropriate.
These tests indicate whether the spatial dependence in the OLS models is
more likely to be the result of place-based processes (a common exposure
to a causal factor not in the model) or propagation-based processes (diffu-
sion).31
In the former scenario, a spatial error model is the most appropriate
spatial specification. In the latter case, a spatial lag model is more appropriate.
For each OLS model we estimate, we run the spatial model recommended
by the Lagrange Multiplier statistics. We specify a spatial weights
matrix based on the principle of “queen contiguity.” In other words, each
electoral district is considered to be a neighbor to all electoral districts
with which it shares any length of its border. Our results, presented below,
demonstrate the extent to which modeling spatial dependence in areal data
can significantly improve model fit. In the final section, we explore the spatial
dynamics of our results and their potential implications in more detail.
Table 1 below first presents results for an OLS model of the effects of
cleavages and legacies on CDU support along with the spatial regression
diagnostics. In this first model, the class and religious cleavage variables do
not have a significant effect on CDU vote, though the relative size of the
migrant population has a very significant and negative effect on support for
the CDU. The legacy variable was also (weakly) statistically significant, suggesting
that political configurations at the time of the transition augur well
••• 8 •••
Steven Wuhs and Eric McLaughlin
for the CDU long term. Voter turnout was a strong negative predictor of CDU
support, which was not unexpected given the defensive position the party
and Chancellor Merkel were in during the campaign.
Table 1: 2017 Federal Elections (Electoral District)
Dependent Variable: Percent CDU Vote
The Lagrange Multiplier tests for both error (place-based) and lag (diffusion
or propagation-based) are statistically significant, indicating that indeed spatial
effects are at play. In an effort to select the most appropriate corrective
model specification when both LM tests are statistically significant, Anselin
••• 9 •••
Explaining Germany’s Electoral Geography
IndependentVariable Cleavage, Legacy, and Migration Model Spatial Lag Model
Constant .000 -.047 ^
(.085) (.063)
Agriculture Insurance .064 .030
(.118) (.086)
Industry Insurance -.127 -.076
(.114) (.084)
% Not Catholic or Protestant -.182 -.168
(.128) (.094)
CDU Membership (1990) .183 ^ .132
(.102) (.076)
% Migrant Population -.422 * -.403 ***
(.156) (.115)
% Turnout -.387 ** -.218 *
(.130) (.102)
% Unemployment .115 .057
(.108) (.079)
Spatial Lag Term (ρ) _ _ _ .538 ***
Moran’s I (error) 3.986 ***
Lagrange Multiplier (lag) 19.557 ***
Robust LM (lag) 10.862 ***
Lagrange Multiplier (error) 9.093 **
Robust LM (error) .398
N = 49 49
R-Squared (Adjusted or Pseudo) .644 .806
^ p < .10
* p < .05
** p < .01
*** P < .001
and Rey suggest looking next to the robust forms of these two test statis-
tics.32
Here, the robust form of the error model is not significant, suggesting
that the spatial lag model is appropriate.
Table 2: 2017 Federal Elections (Electoral District)
Dependent Variable: Percent SPD Vote
Respecifying the model to include a spatial lag produces only a small
change in substantive results, in that the legacy variable is no longer significant
at even 90 percent confidence. Only migration background and bloc
••• 10 •••
Steven Wuhs and Eric McLaughlin
IndependentVariable Cleavage, Legacy, and Migration Model Spatial Lag Model
Constant .000 .004
(.096) (.058)
Agriculture Insurance .619 *** .218 *
(.133) (.088)
Industry Insurance -.198 -.033
(.129) (.078)
% Not Catholic or Protestant -.075 -.036
(.145) (.087)
CDU Membership (1990) .115 .010
(.115) (.071)
% Migrant Population .707 *** .430 ***
(.177) (.110)
% Turnout -.658 *** -.348 ***
(.146) (.091)
% Unemployment -.425 ** -.264 ***
(.122) (.073)
Spatial Lag Term (ρ) _ _ _ .701 ***
(.086)
Moran’s I (error) 3.803 ***
Lagrange Multiplier (lag) 26.066 ***
Robust LM (lag) 19.249 ***
Lagrange Multiplier (error) 8.131 **
Robust LM (error) 1.314
N = 49 49
R-Squared (Adjusted or Pseudo) .546 .835
^ p < .10
* p < .05
** p < .01
*** P < .001
CDU membership are significant predictors, both in the same direction as in
the OLS model. The strong propagation effects in the model are captured by
the ρ parameter, which is positive and statistically significant. The inclusion
of the spatial lag term increases the explanatory power of the model, raising
the adjusted R-square from .644 to .806.
In our second set of models, we examine the capacity of our independent
variables to account for SPD support. We expect to encounter inverse
relationships in these data: weaker support in more agricultural districts,
stronger in industrial ones, greater support in more secular regions, and a
negative effect for CDU membership. In terms of the migration variable,
while we do not have an a priori expectation, given the strongly negative
effect of higher density of immigrant populations on the CDU we expect that
a positive relationship may hold for the SPD.
The results in Table 3 confounded many of those expectations. The OLS
model shows positive and statistically significant effects for agricultural
employment and migrant population and negative effects for turnout and
unemployment. One might explain the agricultural effect by reflecting on
standards of living in rural areas in the east, many of which are significantly
depressed; support for such an explanation would require additional
research. And support for the CDU in districts with higher percentage of
immigrants and/or second-generation Germans likely relates to urbanization—those
populations and SPD support are strongly anchored in Germany’s
major cities. Turnout and unemployment both have negative and
statistically significant effects on SPD support. These variables speak to the
challenging position of the SPD in contemporary German politics. As the
country’s historical mass party, it should be helped by turnout, not hurt by
it, and would have prided itself on support for the unemployed. Yet, recent
elections have seen a secular decline in SPD support, from near 40 percent
nationwide in the 2002 federal elections to just 20 percent in 2017. It is not
succeeding in offering a promise to Germany’s working class.
Both LM tests (error and lag) are statistically significant. In comparing the
two robust versions of the tests, however, we again observe that the robust
test for spatial lag is statistically significant, while the robust test for spatial
error is not. Once again following Anselin and Rey, we opt to model the
spatial dependence in this model using a spatial lag specification, which is
reported in the second column. In terms of significant predictors and directions
of effects, nothing changes here other than the relative magnitudes of
those effects. As before, the spatial lag term is positive and statistically significant
and the pseudo R-squared for the model is significantly higher than
the adjusted R-squared in the non-spatial model (.835 vs. .546).
••• 11 •••
Explaining Germany’s Electoral Geography
Table 3: 2017 Federal Elections (Electoral District)
Dependent Variable: Percent Left Party Vote
Our third model considers the performance of the Left Party, which won
9.2 percent of the vote nationally. As noted earlier, we did not have clear
expectations for the bases of Left Party support because while it sits on the
left, as the successor party to the East German Socialist Unity Party, it has a
complicated political identity. We did note, though, that compared with the
other three parties, it does have a fairly comprehensive organizational
••• 12 •••
Steven Wuhs and Eric McLaughlin
IndependentVariable Cleavage and Legacy Model Spatial Error Model
Constant .000 .084
(.105) (.264)
Agriculture Insurance -.032 .315 *
(.145) (.141)
Industry Insurance -.456 ** -.521 ***
(.141) (.106)
% Not Catholic or Protestant .133 .199
(.158) (.123)
CDU Membership (1990) -.052 -.074
(.126) (.095)
% Migrant Population .250 .391 **
(.193) (.150)
% Turnout -.198 -.008
(.160) (.150)
% Unemployment -.024 .048
(.133) (.096)
Spatial Error Term (λ) _ _ _ .693 ***
(.111)
Moran’s I (error) 3.352 ***
Lagrange Multiplier (lag) 3.417 ^
Robust LM (lag) .311
Lagrange Multiplier (error) 5.993 *
Robust LM (error) 2.887 ^
N = 49 49
R-Squared (Adjusted or Pseudo) .458 .682^
p < .10
* p < .05
** p < .01
*** P < .001
structure across the region because of its history. In our OLS model, industry
insurance is the only significant predictor, with a negative coefficient. In
areas with higher concentrations of industrial workers, the Left Party does
worse. This is consistent with a different legacy argument: that industrial
workers captured for decades by the SED would be reluctant to support the
successor. That logic certainly held in the first decades after reunification,
and we see some evidence for that here.
Unlike in the models for the CDU and SPD, the diagnostic statistics for
this model suggest that a spatial error (common exposure) specification
would be the most appropriate way to model the spatial dependence lurking
in these data, since only the LM test for error is significant at the 95 percent
confidence level. The second column of Table 3 presents the results of
the spatial error model. In this model, in which the spatial dependence is
modeled as the result of the areal units’ common exposure to some factor
or factors not in the model, both of the economic cleavage variables are
significant. Agricultural insurance positively, though weakly, predicts support
for the Left Party, whereas industry insurance remains a negative predictor.
The migration variable is positive and statistically significant, so
districts with higher concentrations of migrants were more likely to support
the Left Party—the same as the SPD and the inverse of the CDU, suggesting
that the campaign’s focus on migration did exert an influence on
voting. Lambda, the spatial error term, is positive and significant as
expected, and respecifying the model in this way boosted its R-squared
value from .458 to .682.
Our fourth and final model consider the performance of the Alternative
for Germany, which surged to 12.6 percent of the national vote in the 2017
federal elections. How well does our model account for the AfD’s performance,
and what is the relative power of cleavage, legacy, and spatial factors
in accounting for its success?
Our OLS results in the first column of Table 4 demonstrates that social
cleavages shaped AfD performance. Districts with higher levels of agricultural
insurance were less likely to support the AfD, while areas with more
industrial workers were more likely to. Districts with higher proportions of
migrants were far less likely to vote AfD, and voter turnout was a strong
and positive predictor of AfD success.
The LM test statistics suggest that a spatial lag specification is appropriate.
The second column shows the results for the spatial lag model, which as in
the other cases has significantly improved explanatory power over the OLS
model. In addition to the powerful spatial lag term, cleavage effects are
muted while migration and turnout both remain significant.
••• 13 •••
Explaining Germany’s Electoral Geography
Table 4: 2017 Federal Elections (Electoral District)
Dependent Variable: Percent AfD Vote
Discussion
Taken together, these models tell us a lot about support for the CDU and its
competitors in Germany’s eastern regions. For the governing party, variables
related to the economy were not significant predictors of the vote for
the past federal election; nor was the religion or religiosity of the electorate.
In fact, the district-level sociodemographic variable that best accounted for
CDU vote share was the immigrant population share. Given the centrality of
••• 14 •••
Steven Wuhs and Eric McLaughlin
IndependentVariable Cleavage, Legacy, and Migration Model Spatial Lag Model
Constant .000 -.091
(.103) (.080)
Agriculture Insurance -.336 * -.029
(.142) (.094)
Industry Insurance .315 * .151 ^
(.138) (.085)
% Not Catholic or Protestant .083 .064
(.155) (.094)
CDU Membership (1990) -.163 -.029
(.123) (.076)
% Migrant Population -.687 *** -.466 ***
(.189) (.117)
% Turnout .498 ** .194 *
(.156) (.095)
% Unemployment .137 .056
(.130) (.079)
Spatial Lag Term (ρ) _ _ _ .767 ***
Moran’s I (error) 4.544 ***
Lagrange Multiplier (lag) 22.053 ***
Robust LM (lag) 9.821 **
Lagrange Multiplier (error) 12.355 ***
Robust LM (error) .123
N = 49 49
R-Squared (Adjusted or Pseudo) .482 .805^
p < .10
* p < .05
** p < .01
*** P < .001
the migrant issue in the 2017 contest, it is not surprising that Merkel’s party
was punished in areas with higher proportions of migrants. Those weak
social foundations also held in the case of the SPD, where only agricultural
employment significantly affected its vote share in 2017. Germany’s two
major parties showed little evidence of institutionalized bases of support in
the electorate. In startling contrast, one of the most compelling findings
here was that the upstart AfD had capitalized on social divisions in that contest,
a particularly concerning harbinger for the mainstream parties.
Our results show statistically significant and negative effects of migrant
populations on CDU and AfD support, and statistically significant and positive
effects on SPD and Left Party support. These are not straightforward
effects to interpret, given the expectation that migration would affect the
CDU and AfD differently, and that it would result in higher levels of AfD
support. We suspect that this variable may in fact be tapping into urbanization-related
dynamics. The AfD’s support is lower in districts with higher
concentrations of historical migrants (i.e., urban areas), while in areas with
lower numbers of historical migrants (i.e., rural areas), it performs better.
That the CDU fares better in more rural districts is fairly conventional logic
in German politics. Likewise, the SPD’s and Left Party’s support in areas
with higher numbers of migrants is not surprising in light of an urban interpretation.
In the end, despite how robust the effects of this variable are, we
cannot say conclusively whether the migrant populations themselves play a
role here, or if the effect is a spurious one.
Our legacy variable, bloc CDU membership, is consistently weak across
our models, influencing only the CDU vote in 2017. Recall that we argue that
this variable represents a measure of the anti-regime presence during the
SED period. Localities with higher bloc-CDU membership were typically
either bastions of Catholic culture (like the Eichsfeld region), peripheral
areas in the GDR that were less subject to direct state control (as compared
to Berlin and Brandenburg, for example, where the regime was based), or
sites of protest through the 1980s (as in the contemporary state of Saxony).
In these cases, we suggest, localized cultures of resistance were created that
may continue to mark party politics today, at least the CDU.33
Finally, the models show a consistent spatial effect in support for all of
the political parties we examined. Whether utilizing a spatial error model
(to account for place-based effect)) or a spatial lag model (addressing propagation
effects), the spatial terms in our models were uniformly statistically
significant and improved model fit. We also observed how modeling the
spatial structure of our data resulted in different inferences about the relative
import of our independent variables.
••• 15 •••
Explaining Germany’s Electoral Geography
Why would the inclusion of these spatial factors have such a pronounced
effect on model fit? The short answer is simply that party building as a
process is highly territorialized, governed by parties’ goals of winning office.
While electoral districts (our unit of analysis) are one site of competition for
parties, party leaders also focus their organizational energies on Land- and
Kreis-level party building as well as their municipal presence.34
To the extent
that those other party-building efforts are affecting how German voters mark
their ballots, they manifest as spatial effects in our model. Interviews with
party leaders suggest other spatial factors that might drive how eastern Germans
vote: proximity to the border with western Germany or the Czech and
Polish borders; the distance from Berlin; and even the structure of media
markets through which parties and politicians reach out to voters.35
Figure 1: Residuals from OLS Model Predicting 2017 CDU Vote Share
(upper) and SPD Vote Share (lower), 49 Electoral Districts)
Lighter = Positive Residuals (model underpredicted CDU or SPD vote)
Darker = Negative Residuals (model overpredicted CDU or SPD vote)
••• 16 •••
Steven Wuhs and Eric McLaughlin
The simple exercise of mapping the residuals from of our spatial models
accomplishes two related goals. First, it brings the territoriality of party support
into sharp visual relief. Secondly, it can serve as the foundation for the
generation of conjectural hypotheses as to the potential sources of that territoriality.
Imke Harbers and Matthew Ingram note that while quantitative
spatial error and spatial lag models identify the type of spatial dependence
that exists in a dataset, they do not tell their users anything about the
sources of common exposure (as in a spatial error model) or the nature of
the vectors of diffusion that produce propagation effects across adjacent
units (as in a spatial lag model). They argue that quantitative analysis such
as those presented here can be used to set the agenda for future research.36
Figure 1 present choropleth maps of regression residuals, with lighter
shades representing positive regression residuals (districts where observed
support for a party in 2017 was higher than predicted by the regressors in
the model) and darker shades representing negative residuals (districts
where observed support for a party in 2017 was lower than predicted by the
regressors in the model). Below, we use these maps as the basis to offer
some thoughts about the sources of spatial dependence in the 2017 election
for the mainstream parties, the CDU and SPD, and the AfD because of its
surge in the election.37
In these first maps of residuals, both images show clear evidence of statelevel
spatial effects: in the state of Mecklenburg-West Pomerania (Merkel’s
home state) the CDU substantially outperforms the expectations of the
model in all electoral districts, while in much of Brandenburg, where the
CDU is considerably weaker organizationally (the upper map), it does poorer
than the model expects. The SPD’s performance in Mecklenburg-West
Pomerania and Brandenburg mirrors that: overperforming in its western
bastion of Brandenburg and underperforming in the northern state. A statelevel
story is less evident in the residual analysis for Saxony, SaxonyAnhalt,
and Thuringia, however, so those state-level effects are not
hard-wired. That is surprising in the case of Saxony given the hegemony of
the CDU for almost all of the postreunification period there, though our consideration
of the AfD below offers some insights on that.
There does appear to be evidence of border effects: the southern corridor
of Thuringia (bordering Bavaria) and the Czech and Polish border corridors
in Saxony and Brandenburg are areas where the CDU did not perform as well
as expected by the model. That is, in areas that are vulnerable to migrant
flows or perhaps perceived as such, traditional models of voting do not work
as well, though not in a predictable direction since the CDU both outperforms
or underperforms the model’s expectations. We see some related evidence
••• 17 •••
Explaining Germany’s Electoral Geography
for the SPD in the eastern portions of Saxony and Mecklenburg-West Pomerania,
near the Czech and Polish borders. While we would not expect strong
SPD support in a state where the CDU has been so strong, it bears mentioning
that neither of the region’s dominant parties performed well in those zones.
Again, the AfD’s performance offers some potential insights.
There are also some isolated pockets worth noting, especially in the case
of the CDU. In the Eichsfeld region in the north of Thuringia and the Erzgebirge
region of Saxony, the CDU showed considerable electoral resilience by
strongly outperforming the expectations of our model. Those regions are
particular historical strongholds for the party, though, so the resilience is
not too surprising.38
Interestingly, the overperformance of the CDU here is
not accounted for by either the CDU’s transition-era membership or the religious
variable we test in the model, even though both are regions considered
more religious and in the Eichsfeld’s case had a strong bloc CDU
organization throughout the GDR period.
Finally, Figure 2 presents the residuals from our analysis of AfD support.
Figure 2: Residuals from OLS Model Predicting 2017 AfD Vote Share, 49
Electoral Districts)
Lighter = Positive Residuals (model underpredicted AfD vote)
Darker = Negative Residuals (model overpredicted AfD vote)
The areas with larger positive residuals (those cases where the model
underpredicted AfD support) are pockets toward the eastern half of the
map, with strong AfD regions near Dresden and the border region of Gör•••
18 •••
Steven Wuhs and Eric McLaughlin
litz. We also see comparatively strong AfD performance both in regions
recently dominated by the CDU (eastern Saxony, eastern Mecklenburg-West
Pomerania) and by the SPD (much of Brandenburg). That distribution
should not surprise, as populist right parties often are the beneficiaries of
anti-party sentiment.
Negative residuals are scattered across the region, but there are concentrations
in western Mecklenburg-West Pomerania, in Thuringia, in western
Saxony, and in districts surrounding Berlin. There is also evidence in Figure
2 of effects of localized party building processes. Earlier, we noted that the
CDU champions its resilient organizational structures in Thuringia and the
Erzgebirge region of Saxony; the AfD underperformed relative to the
model in exactly those regions, suggesting that deeper party organizations
may be able to resist populist challengers.
The residual maps suggest that AfD support (like the CDU’s) may be
related to proximity to regional borders. To the east, we observe in Figure 2
that the AfD’s performance is substantially underpredicted by the model—
precisely the inverse of the CDU’s case. Interestingly, in regions bordering
western Germany the AfD underperforms. Taken together, those effects
suggest that the populist right’s performance may have been shaped by perceptions
of economic threat (to the east) and opportunity (to the west), raising
questions about economic divides within Germany’s eastern region.
These examples confirm visually what our statistical analysis above
demonstrated: that alongside structural and historical foundations of Germany’s
electoral geography, there are important underlying spatial dynamics,
and that our explanations are better when we acknowledge and model them.
Conclusion
In recent German federal elections, cleavages structures did not shape
electoral outcomes, at least for the historically dominant parties in the
region, the CDU and SPD. Their negligible effect on vote choice may be a
result of their weakening hold on the German electorate more broadly, or
perhaps more narrowly an effect of the GDR’s totalitarian interlude which
dampened social divisions. Only the radical right upstart AfD’s support seemed
attached to social cleavages, suggesting that the potential remains for a
vote structured by structural divisions in society but also highlighting that
the primary German parties are not drawing on them.
The effect of migrant populations was also mixed across the parties, positively
affecting SPD and Left Party support while undercutting CDU and AfD
••• 19 •••
Explaining Germany’s Electoral Geography
support—though this may have more to do with where migrants have settled
historically (i.e., urban areas in Germany) than the presence of migrants
themselves or the politics around recent migrants in the 2017 campaign. We
also considered how historical legacies shaped voting, but our legacy variable
was significant only for the CDU and only in the traditional OLS model.
Indeed, migration aside, the most robust variable in our OLS models was
introduced as a control: turnout was an important predictor of support for
all but the Left Party (negatively affecting the CDU and SPD, and positively
affecting the AfD).
Unlike most of our other predictors, spatial variables were robust and
powerful determinants of support for the CDU and its competitors, substantially
improving the explanatory power of all four models. The consistency
of effects presented here should signal to scholars of parties and electoral
politics the need to incorporate geography into their analytical frameworks.
Eastern Germany may in fact be less spatialized than other country cases
because of the homogenizing efforts of the SED regime, yet our results show
that across elections the effects of spatial dependence outweigh both cleavage
and historical effects; they are dramatically powerful predictors of CDU,
SPD, Left Party, and AfD support. If that is the case in eastern Germany,
controlling for structural and historical factors, it is likely to be an even
more powerful predictor of political outcomes in western Germany and
other countries with pronounced traditions of territorial politics like Belgium,
Spain, and the United Kingdom.
STEVEN WUHS is Professor of Political Science at the University of Redlands.
His work examines democratization, party development, and electoral
geography in Germany, Mexico, and the United States. Between 2011
and 2014 he was Fellow of the Alexander von Humboldt Foundation at the
Technical University of Dresden.
ERIC MCLAUGHLIN is Associate Professor of Political Science at the University
of Redlands. His research and teaching interests include statistical
methodology as well as party politics, elections, and the politics of ethnicity
and nationalism, particularly in sub-Saharan Africa.
••• 20 •••
Steven Wuhs and Eric McLaughlin
Notes
1. Peter C. Ordeshook and Olga V. Shvetsova, “Ethnic heterogeneity, district magnitude,
and the number of parties,” American Journal of Political Science 38, no. 1 (1994): 100-123.
2. Seymour Martin Lipset and Stein Rokkan, “Cleavage Structures, Party Systems, and
Voter Alignments: An Introduction” in Party Systems and Voter Alignments: Cross-National
Perspectives, ed. Seymour Martin Lipset and Stein Rokkan (New York, 1967); Stein
Rokkan, State formation, nation-building, and mass politics in Europe: the theory of Stein
Rokkan: based on his collected works (New York, 1999); Daniele Caramani, The Nationalization
of Politics: The Formation of National Electorates and Party Systems in Western Europe (New
York, 1994).
3. Russell J. Dalton and Martin P. Wattenberg, eds., Parties without partisans: Political change
in advanced industrial democracies (Oxford University Press on Demand, 2002); Hanspeter
Kriesi, “The Transformation of Cleavage Politics: The 1997 Stein Rokkan Lecture,” European
Journal of Political Research 33 (1998): 165-185; Mattei Dogan, “Erosion of class voting
and of the religious vote in Western Europe.” International Social Science Journal 47, no. 4
(1995): 525; Russell J. Dalton, et al. Electoral change in advanced industrial democracies:
Realignment or dealignment? (Princeton, 1984); Herbert Kitschelt, The Logics of Party Formation:
Ecological Politics in Belgium and West Germany (Ithaca, 1989); Herbert Kitschelt, Postcommunist
party systems: competition, representation, and inter-party cooperation (New York,
1999); Pippa Norris, Radical right: Voters and parties in the electoral market (New York, 2005).
4. Martin Elff, “Social Structure and Electoral Behavior in Comparative Perspective: The
Decline of Social Cleavages in Western Europe Revisited,” Perspectives on Politics 5 no. 2
(2007): 284.
5. Robin E. Best, “The declining electoral relevance of traditional cleavage groups,” European
Political Science Review 3 no. 2 (2011): 279-300.
6. Fraser Duncan, “Preaching to the converted? Christian Democratic voting in six west
European Countries,” Party Politics 21 no. 4 (2015): 577-590.
7. Kenneth M. Roberts, Changing Course in Latin America (Cambridge, 2015).
8. Anna M. Grzymala-Busse, Redeeming the Communist Past: The Regeneration of Communist
Parties in East Central Europe (New York, 2002).
9. Ibid; see also Peter Mair, Party system change: approaches and interpretations (New York,
1997); Dan Slater and Joseph Wong, “The Strength to Concede: Ruling Parties and
Democratization in Developmental Asia,” Perspectives on Politics 11 no. 3 (2013): 717-733.
10. As parties shape their fortunes through key decisions they make in particular historical
contexts, voters too habituate themselves to particular kinds of behaviors. While the present
article does not consider individual-level political behavior, there is a notable connection
between system-level lock-in phenomena discussed by historical institutionalists
and its micro-level manifestations. Voters are habituated to the act of voting, and there is
further evidence that that habituation extends to loyalty to particular political parties—in
fact crystallizing over time into very stable behaviors approaching “lock-in.” See John H.
Aldrich, et al., “Turnout as a Habit,” Political Behavior 33 no. 4 (2011): 535-563; Ronald
Alfaro-Redondo, “Lifecycle Changes and the Activation of Habitual Voting: The Case of
Costa Rica,” Electoral Studies 35 (2014): 188-199; Daniel de Kadt, “Voting Then, Voting
Now: The Long-Term Consequences of Participation in South Africa’s First Democratic
Election,” The Journal of Politics 79 no. 2 (2017): 670-687; Raul Gomez, “All that You
Can(not) Leave Behind: Habituation and Vote Loyalty in the Netherlands,” Journal of
Elections, Public Opinion, and Parties 23 no. 2 (2013): 134-153.
11. Waldo R. Tobler, “A computer movie simulating urban growth in the Detroit region,”
Economic Geography 46, sup1 (1970): 234-240; Patrick Doreian, “Linear models with spatially
distributed data: spatial disturbances or spatial effects?” Sociological Methods &
Research 9, no. 1 (1980): 29-60.
12. David Darmofal, Spatial analysis for the social sciences (New York, 2015).
••• 21 •••
Explaining Germany’s Electoral Geography
13. Imke Harbers, “Spatial effects and party nationalization: The geography of partisan support
in Mexico,” Electoral Studies 47 (2017): 55-66.
14. Interviews with CDU member and former Saxony Minister-President Georg Milbradt,
September 2012; Konrad Adenauer Foundation Educational Researcher Joachim Klose,
June 2009; and CDU leader of Saxony-Anhalt Thomas Webels, December 2012.
15. Klose Interview, ibid.; CDU legislator in Saxony State Legislature Steffen Flath, December
2012; Konrad Adenauer Foundation-Brandenburg Researcher Dr. Stephan Raabe,
December 2012.
16. Joshua D. Potter and Santiago Olivella, “Electoral strategy in geographic space: Accounting
for spatial proximity in district-level party competition,” Electoral Studies 40 (2015): 76-
86; Harbers (see note 13).
17. Statistisches Amt Mecklenburg-West Pomerania, ed. Statistiches Jahrbuch Mecklenburg-West
Pomerania (Schwerin, 1991-2012); Amt für Statistik Berlin-Brandenburg, ed. Statistisches
Jahrbuch Brandenburg Berlin (Berlin, 1991-2012); Statistisches Landesamt des Freistaates
Saxony, ed. Statistiches Jahrbuch Saxony (Kamenz, 1991-2012); Thüringer Landesamt für
Statistik, ed. Statistiches Jahrbuch Thuringia (Erfurt, 1991-2012). The state of Saxony-Anhalt
is excluded here because of the lack of comparable data.
18. Anteil der evangelischen Bevölkerung auf Landkreisebene nach Ergebnissen des Zensus
2011 in Deutschland. That decline is uneven, notably. In the state of Thuringia, for example,
Catholic parishes outside the Eichsfeld region (a particularly dense Catholic pocket
along the old GDR-FRG border) have lost nearly double the members of the parishes
within the region.
19. Deutsche Gewerkschaftsbund, Geschäftsbericht des Bundesvorstandes des Deutschen Gewerkschaftsbundes
1990 bis 1993 (Düsseldorf, 1994). More recent data from the EU confirm
that overall unionization rates in the eastern states are markedly lower than in the west.
See https://www.eurofound.europa.eu/observatories/eurwork/articles/industrial-relations/
germany-revival-of-german-trade-unions, accessed 13 July 2018.
20. Jennifer Yoder, From East Germans to Germans? The New Postcommunist Elites (Durham,
1999); Louise Davidson-Schmich, Becoming Party Politicians: Eastern German State Legislatures
in the Decades following Democratization (Notre Dame, 2006).
21. William Mishler and Richard Rose, “What are the origins of political trust? Testing institutional
and cultural theories in post-communist societies” Comparative Political Studies 34
no. 1 (2001): 30-62; Mariano Torcal, “The Decline of Political Trust in Spain and Portugal:
Economic Performance or Political Responsiveness?” American Behavioral Scientist 58
no. 12 (2014): 1542-1567.
22. Russell J. Dalton and Wilhelm Bürklin, “Wähler als Wandervögel: Dealignment and the
German Voter,” German Politics and Society 21 no. 1 (2003): 64.
23. The results of county level analyses are available from the authors as well. The state of
Mecklenburg-West Pomerania had to be excluded from that analysis because of the timing
of its reform of county boundaries (Kreisgebietsreform) in 2011 just after the 2010 German
census.
24. Heather Stoll, “Social cleavages and the number of parties: How the measures you
choose affect the answers you get,” Comparative Political Studies 41 no. 11 (2008): 1443-1453.
25. See https://www.bundeswahlleiter.de/bundestagswahlen/2017/strukturdaten/bund-99.
html, accessed 18 April 2019.
26. Data available from the authors upon request.
27. Steven T. Wuhs, “Paths and Places of Party Formation: The Post-Unification Development
of Germany’s CDU,” Comparative Politics 49 no. 1 (2016): 43-62.
28. Karsten Grabow, “The Re-Emergence of the Cadre Party? Organizational Patterns of
Christian and Social Democrats in Unified Germany,” Party Politics, 7 (2001): 23-43;
Dieter Segert, “The East German CDU: An Historical or a Post-Communist Party?” Party
Politics 1 (1995): 589-598.
29. Susanne Lohmann, “The Dynamics of Informational Cascades: The Monday Demonstrations
in Leipzig, East Germany, 1989–1991,” World Politics 47 (1994): 42-101.
••• 22 •••
Steven Wuhs and Eric McLaughlin
30. Luc Anselin and Sergio J. Rey, Modern spatial econometrics in practice (Chicago, 2014).
31. Harbers (see note 13).
32. Anselin and Rey (see note 30).
33. These dynamics resemble those identified by Jason Wittenberg, who argued that Christian
churches’ resistance to the socialist regime’s secularization efforts there powerfully
shaped the electoral preferences of Hungarian voters after the transition there. Jason Wittenberg,
Crucibles of Political Loyalty: Church Institutions and Electoral Continuity in Hungary
(New York, 2012).
34. Wuhs (see note 27).
35. Interviews with Kurt Biedenkopf, CDU leader and former Saxon minister-president, September
2012; Lorenz Caffier, Deputy Minister President and CDU Member, MecklenburgWest
Pomerania, October 2012; Thomas Webels, Landesvorsitzender, CDU-Thuringia,
December 2012; Georg Milbradt, CDU-Saxony leader and former Minister-President, September
2012; Eckhardt Rehberg, CDU -Mecklenburg-West Pomerania leader and Landtag
member, September 2012.
36. Imke Harbers and Matthew C. Ingram, “Geo-nested analysis: Mixed methods research
with spatially dependent data,” Political Analysis 25 (2017): 289-307.
37. We exclude the Left Party from this portion of the analysis. Its performance is less
strongly spatialized than for the other parties, evident in the comparatively low Moran’s I
reported earlier. That makes sense since unlike the CDU and SPD, which had to build
party organizations following reunification, as the successor party to the Socialist Unity
Party (SED), the Left Party inherited a territorial structure. Mapped residuals for the
party’s performance in 2017 also showed less clear geographic patterns and smaller overall
residuals. Data available from the authors upon request.
38. Interviews with Flath; Manfred Grund, CDU Federal Legislator from Eichsfeld, June 2013;
CDU leaders from Eichsfeld Rolf Berend and Kerstin Sommerfeld, July 2013.
••• 23 •••
Explaining Germany’s Electoral Geography
Reproduced with permission of copyright owner.
Further reproduction prohibited without permission.