do political parties matter? - evidence from german municipalities · 2019-01-10 · do political...
TRANSCRIPT
Do Political Parties Matter? - Evidence from
German Municipalities∗
Nadine Riedel† Martin Simmler‡ Christian Wittrock§
Abstract
This paper tests whether partisanship of local councils affects the level and composi-
tion of public spending by West German municipalities between 1994 and 2006. Our
identification strategy exploits changes in the party with the absolute majority in the
local council in combination with a matching strategy to address potential selection
into treatment. We find evidence for strong partisan effects: In communities with left-
wing councils, spending on ’people-oriented’ public goods, especially recreation goods
and social services, is significantly higher and spending on infrastructure public goods
is significantly lower than in communities with right-wing dominated councils.
.
Keywords: partisan effects, local governments, public spending
JEL Classification: H7, H4, R5
∗We are grateful to participants of the Congress of the German Economic Association in Vienna
and of the Annual Conference of the International Institute of Public Finance in Tokyo for helpful
comments and suggestions. We gratefully acknowledge financial support from the German Research
Foundation (Simmler: SI 2050/1-1 and Riedel: SI 2491/2-1).†Ruhr-University Bochum. E-mail: [email protected].‡University of Oxford Centre for Business Taxation. E-mail: [email protected].§RGS Econ, TU Dortmund and U Bochum. E-mail: [email protected]
1 Introduction
Do political parties matter? A growing body of empirical research suggests that the
answer to this question is ‘yes’, presenting evidence that political partisanship affects
various policy outcomes, including per-capita spending, spending composition and tax
rate choices (see e.g. Besley and Case (2003), Lee et al. (2004), Potrafke (2011a) and
Potrafke (2011b)).1 Existing studies on partisan effects are, however, largely restricted
to the state and federal level. Studies for localities are scarce, despite the fact that the
national and state results may not carry over to the local level, where interjurisdictional
mobility is high and Tiebout sorting into homogenous local units or competition for
mobile tax bases may limit the scope for partisan politics.
There are three recent exceptions. Ferreira and Gyourko (2009) study the link be-
tween partisanship and public spending of US cities, presenting evidence that partisan
effects are largely absent at the US city level. Freier and Odendahl (2015), in turn, pro-
vide evidence that the party composition of the local council and the party affiliation of
the mayor influence jurisdictions’ local business tax rate choices in the German state of
Bavaria. Sole-Olle and Viladecans-Marsal (2013) show that local left-wing governments
in Spain are less likely to convert land from rural to urban uses than their right-wing
counterparts.
We add to this literature by empirically testing for effects of partisanship on overall
spending and the composition of spending in German localities, which are the lowest
government tier in Germany. The analysis relies on detailed data on municipality spend-
ing across different spending categories between 1994 and 2006, including e.g. spending
for child and youth care, culture and infrastructure. The data is linked to information
on the party composition of local councils, which are the local legislative bodies that
decide on local tax and spending policies.
Our empirical strategy is a simple fixed effect regression framework using (ln)
overall spending and the spending shares for different sub-categories as dependent
variables. The main explanatory variable is a dummy if the SPD (CDU), the main left-
wing (right-wing) party in Germany, holds the majority of seats in the local council.
To avoid that our variable of interest captures spending differences between councils
with absolute majorities and coalition majorities, we focus our analysis on jurisdictions
with a change in the party that has the absolute majority in the local council.
1Related to this literature is work that tries to explain why political parties take on extreme
positions, see e.g. Alesina (1988), Besley and Coate (1997) and Glaeser et al. (2005)
1
While the simple fixed effect regression approach absorbs potential time-invariant
unobservable confounders, our estimates might still be biased by potential selection
into treatment. To address this, we combine the fixed effect with an entropy balancing
approach. Since the underlying reasons for a selection into treatment may differ between
jurisdictions with a change in the council majority from SPD to CDU compared to
from CDU to SPD, we construct two subsamples of jurisdictions and exploit that we
observe communities for two consecutive election periods. The first subsample only
includes jurisdictions in which we observe a SPD majority in the local council in the
first election period and either a remaining SPD majority in the consecutive election
period or a change to a CDU majority. In the second subsample, we only include
jurisdictions with a CDU council majority in the first election period, which either
remained unchanged in the second election period or changed to a SPD majority.2
Treatment and control communities are matched within these two subsamples. As
the estimation model includes a rich set of socio-economic municipality controls, our
matching variables focus on jurisdictions’ overall spending, spending composition and
local tax rates (as well as their changes) prior to the second period election. To ensure
covariate-balancing (see, e.g., Hainmueller (2012)), we employ entropy-balancing and
match on the first two moments of the distribution of the matching variables as well
as on the mean of federal state dummies.
The empirical approach hence resembles a difference-in-differences estimator that
compares changes in the level and composition of community spending between lo-
calities that are observationally identical prior to the second period election but then
do and do not experience a change in the council majority in the second election pe-
riod. Note that the model includes a broad set of control variables. Coupled with the
matching strategy, this implies that the empirical identification approach allows for dif-
ferences in underlying spending (composition) trends between treatment and control
jurisdictions that root in observed community characteristics or in unobserved factors
that correlate with localities’ pre-election spending characteristics that are used as
matching variables in the matching strategy.
Our results do point to quantitatively important partisan effects. When the council
majority changes from CDU to SPD (i.e. right-wing to left-wing), we observe a signif-
icant change in the composition of community spending: Spending for publics goods
2To increase the matching quality, our baseline estimates furthermore add communities with a
left-wing (right-wing) majority in the local council in both election periods to the control group in
the first (second) subsample.
2
and services that are targeted towards people (mainly spending for recreational goods
and services and social spending) increases, while spending for infrastructure goods
declines. The level of overall locality spending, in turn, remains unchanged. When the
council majority changes from SPD to CDU, our results, in turn, reject partisan effects
if all years of the electoral period are considered. When we focus on later years in
the election period, symmetric effects emerge in the sense that CDU-dominated local
councils spend more on infrastructure and less on ’people-oriented’ public goods and
services, irrespective of whether council majorities change from SPD to CDU or from
CDU to SPD. The effect size is with 2%-points or 10 to 15% substantial.
In the first post-election years, we in turn observe that both, CDU and SPD, spend
more on people goods and less on infrastructure goods when they gain power and take
over the majority of the local council seats compared to communities where the dom-
inant party in the council remains unchanged. One potential theoretical explanation
for this effect is that opposition parties, in electoral campaigns, have to differentiate
from incumbent policies to sharpen their profile and gain medial visibility. Opposi-
tion parties might find it attractive to promise and, if elected into power, eventually
implement spending policies that put more emphasis on consumptive spending (like
spending for ’people-oriented’ goods and services) than on investment spending (like
infrastructure spending) as the latter goes down better with short-sighted voters. On
top of that, the observed pattern might reflect responses to incumbents’ pre-election
spending choices, which tend to be characterized by increased investment spending
in pre-election years and constant overall spending (see e.g. Kneebone and McKen-
zie (2001), Gonzalez (2002), Brender (2003), Khemani (2004) and Drazen and Eslava
(2010)). Our results are consistent with the notion that parties that gain legislative
power reverse these spending trends (more strongly than incumbents) after the elec-
tion.
Our paper contributes to the prior literature on partisan effects in several ways.
First, we test for partisan effects at the local level, while much of the literature focuses
on state and federal governments and legislative bodies. Existing papers for the local
level, moreover, largely ignore government spending policies and, if they do not, the
spending categories assessed tend to be broad (Ferreira and Gyourko (2009)). We, in
turn, can draw on detailed information on the level and composition of government
spending. While we, similar to Ferreira and Gyourko (2009), do not observe partisan
effects in overall government spending or spending for general public goods, our find-
ings point to partisan effects in detailed spending sub-categories, namely recreational
3
spending and social spending (which dominate the ’people-oriented’ public goods and
service category defined in this paper) and spending for infrastructure, in particular
streets. Secondly, our results stress that tests for partisan effects based on changes
in council majorities are complicated by the fact that opposition parties that come
into power choose spending policies that systematically deviate from spending pat-
terns when majorities remain unchanged (irrespective of whether the change is from
left to right-wing majorities or vice versa). We consider this to be an interesting finding
in its own right. Policy effects of leadership changes have so far received only limited
attention in the empirical literature, with Brender and Drazen (2013) being a notable
exception.3
The remainder of the article is as follows. Section 2 describes the institutional
background and the data. The methodology is outlined in section 3 and section 4
presents the results. Finally, section 5 concludes.
2 Institutional Background and Data
Our empirical analysis assesses the impact of partisanship of local council majori-
ties on overall local spending and the composition of municipal spending using West
German localities as a testing ground.4 In the following, we describe the institutional
background and the data set used for the empirical analysis.
2.1 Institutional Background
According to the German constitution, German municipalities have elected legisla-
tive bodies and governments and have the right to solve any local matters autonomously
(Article 28 of the German constitution). Localities generate income mainly from three
3Brender and Drazen (2013) investigate in a panel of 71 countries whether leadership change affects
an index that aggregates changes in government expenditure over several spending categories. They
find that newly elected leaders in developed countries change expenditure composition in the long run
(i.e. after several years). They attribute the absence of a short run effect to ”legislative, bureaucratic,
or special-interest roadblocks to change” and provide some evidence that suggests it might take time
for newly elected leaders to learn how to steer the budget process in their desired direction. Note that
our testing ground considerably differs from Brender and Drazen (2013) as we study local elections
and determine the impact of changes in local council majorities.4We focus on West Germany as data for public goods and service spending is available for a longer
time span in these localities compared to their East German counterparts.
4
sources. Firstly, a fraction of the personal income tax and the value added tax revenue
administered at the federal and state level are distributed to German municipalities
based on fiscal rules. Second, municipalities receive general and special grants by higher
government tiers. Third, localities have two (major) own revenue instruments at hand:
firstly, they autonomously set the local business tax rate, levied on business income
earned within their borders and secondly, they choose the local property tax.5 The
majority of tax revenues from these two sources remains with the locality, only a minor
fraction is redistributed by fiscal equalization schemes.6 Note that the own tax revenue
instruments generate a significant fraction of local income (on average about 20%).7
German municipalities moreover provide various local public goods and services
(PIGS), e.g. related to the construction and maintenance of roads, sewerage, kinder-
gartens and primary schools. Further, municipalities have to provide social benefits to
the unemployed and social welfare recipients. Additionally, public goods and services
related to culture and sport facilities, tourism, and public transport may be provided.
While some expenditures are mandatory, including administration, social security and
financing liabilities, others are optional, including e.g. spending for theaters, youth
centers, the promotion of science, health care, sport and recreation facilities.
Finally note that legislative processes in the local councils are regulated in the
municipal codes of the community’s hosting state. Municipal codes are similar across
states. Most importantly, in all federal states a simple majority of votes in the local
council is required to enact changes in tax and spending policies.
2.2 Data
As described above, the purpose of our analysis is to test for partisan effects on
local government spending. Our analysis relies on rich data for spending of West Ger-
man localities between 1992 and 2006, which is drawn from municipalities’ accounting
information provided in the Jahresrechnungsstatistik. East Germany is disregarded as
spending information is available from the late 1990ies onwards only. The spending
data allows us to construct spending items for detailed and disaggregated expenditure
categories. Note that, although German municipalities operate in a homogenous en-
5Note that Germany localities set two property taxes (A and B). Property Tax A is applied on
land used for agriculture and forestry. Property Tax B is used for any other build-up property.6Municipalities may also levy other minor local taxes (e.g. a ’dog tax’).7Calculations are based on our sample and the year 2006.
5
vironment, their spending responsibilities are influenced by their size and status. To
increase the comparability of our sample municipalities, we hence focus on small and
mid-size cities with an average number of inhabitants over our sample period between
1,000 and 50,000 and also exclude urban cities (kreisfreie Staedte).
We define different variables to capture the size and structure of municipality spend-
ing, namely (ln) overall real expenditures and (ln) financial expenditures as well as (ln)
voluntary expenditures.8 The voluntary spending measure accounts for the fact that
communities have a number of mandatory spending obligations, which offer no room
for partisan politics. The construction of the voluntary spending measure thus ignores
expenditures that are organized and carried out by higher government tiers, for exam-
ple, at the county level. Second, we exclude spending categories for which the voluntary
dimension tends to be small. See Footnote 9 below for details on the specific spending
categories that enter the voluntary spending variable.
To test for partisan effects on the composition of government, we, in a first step,
account for four relatively broad spending categories. These are (1) spending for gen-
eral public goods, (2) spending for ’people-oriented’ public goods, (3) spending for
culture and (4) spending for infrastructure public goods. Each of the blocs, except of
culture spending, consists of several subcategories. Spending for general public goods
consists of spending for (1.1) public administration, (1.2) public safety, (1.3) commer-
cial enterprises (e.g. own utility or public transport firms) and (1.4) general planning.
Spending for people-oriented public goods consists of spending for (2.1) schools, (2.2)
recreation (as parks and sport facilities) and (2.3) spending for people in need (social
spending). Spending for infrastructure public goods consists of spending for (3.1) roads,
(3.2) public facilities (e.g. sewerage, waste but also public markets) and (3.3) economic
promotion. Testing for partisan effects based on these broad categories offers the ad-
vantage, that estimates are expected to be more precise if the chosen subcategories
are substitutes (what we presume). We will, however, in the following also test for
partisan effects based on individual spending sub-categories. Note, moreover, that in
defining these spending sub-categories, we again define both, comprehensive measures
8Financial expenditures comprise interest payments, debt service as well as taxes and grants trans-
fered to other government tiers.
6
and measures that only comprise voluntary spending within the sub-categories.9 See
Tables (1) and (2) for descriptive statistics. On average municipalities have overall real
expenditures of about 8.96 million Euro and financial expenditures of about 5.89 mil-
lion Euro. The average share of voluntary expenditures to overall real expenditures is
42%. General public spending, spending for ’people-oriented’ goods and infrastructure
spending make up broadly one third of overall spending each. Spending for cultural
goods is, in turn, a small post in communities’ budget.
The sketched spending data is linked to information on the party composition of the
local council obtained from the German Federal Statistical Offices. As will be sketched
in the next section, our empirical identification strategy relies on changes in the party
composition of the local council in the wake of council elections. The timing of council
election varies across German states (as does the term length, which varies between
four and six years). See Table (3) for an overview of election dates and election periods
in each state in our data frame. Note that, on average, we observe three election periods
per state within our data frame.
The party composition of local councils in Germany is shaped by the five main par-
ties, that are also active at the state or federal level, as well as several civil parties that
9Specifically, in all expenditures in the public administration category are defined as voluntary
spending. In term of public safety, spending for policy, public order and the fire department are
classified as voluntary and spending for the environmental office and emergency management as non-
voluntary; For schools, spending on primary and secondary schools as well as spending on schools with
special needs is classified as voluntary, while spending for administration, vocational schools, techni-
cal schools, school transport and other school-related spending is classified as mandatory. In terms of
culture, voluntary spending comprises spending for science and research, museums, collections, the-
aters, concerts, public education and local heritage; non-voluntary spending in that category comprises
spending for administration and churches; In the social spending category, we classify spending for
social welfare centres, nurseries, children and youth work and support for soon-to-be mothers as volun-
tary spending. Non-voluntary in that category comprises spending for administration, social welfare,
support for war victims, support for education, support for young adults and support for families.
In the ’recreation spending’ category, voluntary spending includes spending for sports and own sport
venues, public swimming pools, parks and gardens, other recreation facilities. Non-voluntary spending
in that category comprises spending for administration, hospitals, other health spending. All spending
in the ’general planning’ category (comprising spending for a local building authority and spending for
city planning) is considered to be voluntary. Spending for local streets, county streets, street lightning
and parking facilities is, moreover, considered to be voluntary, while spending for state and country
streets as well as rivers is classified as mandatory. In terms of public facilities, markets and other public
facilities are considered voluntary, while sewerage and waste disposal, slaughterhouses and death care
are mandatory. Moreover, all spending related to public companies is classified as mandatory.
7
Table 1: Descriptive Statistics for Overall Spending and Control Variables
Mean Median Std. Dev.
Overall revenue in million Euro 14.43 6.60 23.02
Real expenditure in million Euro 8.96 3.88 15.03
Financial exp. in million Euro 5.89 2.76 9.24
Overall voluntary public good spending 42.23 42.73 12.91
(as a fraction of overall spending)
Local business tax multiplier 335.80 330.00 33.77
Population in 1000 8.18 3.99 10.96
Share Population under 20 0.23 0.22 0.02
Share Population over 65 0.17 0.17 0.02
Employees in 1000 2.62 1.29 3.49
Unemployment Rate (County) 9.13 8.80 2.37
Debt per capita 942.56 930.39 303.96
Notes: Sample includes all jurisdictions with a SPD or CDU dominated council. Note
that municipalities set a tax multiplier for the local business tax that is reported in the
table. The tax burden is calculated as the product of this multiplier and a base rate
(’Messzahl’) that was 5% for incorporated businesses during our sample period.
Source: Authors’ calculations based on Statistik Lokal and Jahresrechnungsstatistik 1994
to 2006.
focus their activities to the local level. The five main parties are the Christian Demo-
cratic Union and their Bavarian sister party the Christian Social Union in the German
state of Bavaria (in the following, we will refer to both as CDU), the Free Democratic
Party (FDP), the Social Democratic Party (SPD), the Greens (B90/Gruene) and the
party ‘Die Linke’. Following Pappi and Eckstein (1998), these parties can be classified
on a left-wing right-wing scale, where “Die Linke” is on the extreme left and the FDP
on the extreme right. SPD and CDU are the moderate left and right-wing parties. The
Greens are left to the SPD but are, additionally, proponents of environmentally-friendly
policies. Note that this classification scheme cannot be used for the civil parties, whose
programs tend to vary between localities.
To ferret out partisan effects, we will in the following concentrate the analysis on
communities, where the local council is dominated by the large right-wing party, CDU,
or the large left-wing party, SPD, in the observed election periods, or where there is
a switch in the dominating party from SPD to CDU or the other way round. Note
that with ’dominated’, we mean that the considered party holds more than 50% of
the seats in the local council. This design helps us to, firstly, avoid effects related to
8
Table 2: Descriptive Statistics for Spending Composition
Mean expenditure shares for ... ...based on...
in % overall expenditures voluntary expenditures
General public goods 30.73 21.45
Administration 13.82 12.95
Public safety 4.54 4.17
Commercial enterprises 7.75 0.00
General planning 4.62 4.33
’People-oriented’ public goods 33.62 18.12
Schools 10.49 7.50
Recreation 5.47 3.96
Social 17.66 6.67
Culture 1.84 1.23
Infrastructure public goods 33.10 16.96
Traffic 12.53 10.90
Public facilities 18.31 4.44
Economic promotion 2.26 1.62
Notes: Sample includes all jurisdictions with a SPD or CDU dominated council.
Source: Authors’ calculations based on Statistik Lokal and Jahresrechnungsstatistik 1994 to 2006.
Table 3: Elections for the Local Council by State and Year (1993 to 2006)
Year
State 93 94 95 96 97 98 99 00 01 02 03 04 05 06
Schleswig-Holstein X X X
Lower Saxony X X X
North Rhine-Westphalia X X X
Hessia X X X X
Rhineland-Palatine X X X
Baden-Wuerttemberg X X X
Bavaria X X
Saarland X X X
Source: Own data collection.
9
divided governments and political coalitions and, secondly, does not require classifying
local civil parties with varying political positions on the left-right-spectrum. Note,
however, that this also implies that we disregard sample localities, where the local
council is either dominated by a civil party or there is no dominant party at all. In
the end, around 25% of the West German municipalities enter our sample. Around two
thirds of the municipality-year observations is characterized by a local council that is
dominated by the CDU and one third by a local council dominated by the SPD.
We further merged information on socio-economic characteristics of the localities to
our data. These include overall population as well as the age structure of the population,
number of employees, the unemployment rate and debt per capita (the latter two are
available on the county level). Descriptive statistics for the spending variables and the
control variables for communities with an SPD or CDU dominated council respectively
are shown in Table (1). The average municipality has around 8.200 inhabitants and
around 23% (17%) of the population are aged below 20 (above 65).
3 Methodology
Our baseline empirical strategy to assess the impact of partisanship of local coun-
cil majorities on overall spending and spending composition is a simple fixed effect
regression, which reads as follow:
yit = α1Pit + α2Xit + λi + κst + εit. (1)
Our dependent variables are the different measures for the size and structure of local
spending described in the previous section. Our main explanatory variable is Pit, which
is an indicator variable that is one if the ‘SPD’ (CDU) holds the majority (more than
50% of the seat shares) in the local council. The model accounts for municipality fixed
effects λi and state-year fixed effects κst, hence absorbing time constant heterogene-
ity across localities and common shocks to municipality spending in given states over
time. The approach consequently resembles a difference-in-difference strategy, compar-
ing adjustments in the spending of localities with changing majorities in the course of
elections to localities where majorities remained constant. In our sample period, we
observe 59 jurisdictions with a switch in the dominant party in the local council (44
10
from SPD to CDU and 15 from CDU to SPD). While this number of treated observa-
tions is small, the strong sample requirements ensure that the jurisdictions included in
the sample are comparable.
The vector Xit comprises socio-economic control variables for our sample jurisdic-
tions. These include (ln) localities’ population (in 1000), (ln) employed population, age
composition of the jurisdiction as determined by the population shares aged under 20
and above 65 as well as the (ln) unemployment rate and (ln) municipalities debt per
capita (both on the county level). We report robust standard errors that are clustered
at the municipality level.
The Achilles’ heel of the difference-in-differences approach is the common trend
assumption and the assumption that the treatment is exogenous. Thus, treatment and
control group should have behaved similarly in the absence of the treatment and juris-
dictions should not be able to self-select into treatment. In our case, these assumptions
are likely to be violated as the treatment depends on voting behavior in a jurisdiction,
which is influenced by spending policies in the past as well as the jurisdictions’ eco-
nomic environment. Further, municipalities facing different economic environments are
less likely to behave in a similar manner. That this is not only a theoretical concern
can by seen by comparing characteristics of jurisdictions with a SPD dominated coun-
cil and a CDU dominated council (see Columns (1) and (3) in Table (A1) and Table
(A2) in the Appendix). Jurisdictions with SPD majorities tend to be smaller, have
less workers, lower overall expenditures and in particular a lower share of voluntary
spending. To address the potential selection of jurisdictions into the treatment group,
we combine the simple fixed effect framework with a matching strategy.
The most often used matching approach in the literature is propensity score match-
ing (where the propensity score represents the likelihood of being treated). It can be
used to match observations, e.g. to find the closest control unit for every treatment ob-
servation, or to weight observations to create balance between control and treatment
units (see Imbens, 2004, for a review). One particular assumption of the propensity
score approach is that the distribution behind the mean of the matching variables is
the same, as otherwise differences between treatment and control group are not-random
and thus bias the estimated average treatment effect.
One recently proposed approach to overcome the lack of co-variance balancing in
a selection-on-observables framework is entropy-balancing (Hainmueller (2012)). The
main advantage of this method is that covariate balancing is not just assumed but
11
enforced in a constrained, nonlinear estimation approach. The approach obtains weights
for each targeted moment of the balancing/matching variables for treatment and control
group subject to the balancing constraints. The resulting weights can then be used in
a weighted regression.10
Therefore, we apply entropy balancing to balance characteristics of treatment and
control group jurisdictions in the year before the election. Since the underlying reasons
for a change from SPD to CDU majorities in local councils is likely to be different to
a change from CDU to SPD, we construct two subsamples. Each of the two subsample
includes two consecutive election periods of jurisdictions. In the first subsample (Panel
A), we include three types of jurisdictions: (i) jurisdictions which had a CDU majority
in both election periods, (ii) jurisdictions which had a right-wing majority in both
periods and (iii) jurisdictions which had a CDU majority in the first election period
and a SPD majority in the second election period. We do not include jurisdiction in
which the majority changed from right-wing to SPD as this would mean comparing also
divided governments with absolute majorities. We do, however, include jurisdictions
with a right-wing majority in both election periods as this increases the size of the
control group and thus the matching quality. In the second subsample (Panel B), we
include again three types of jurisdictions: (i) jurisdictions with a SPD majority in the
local council in both election periods, (ii) jurisdictions with a left-wing majority in
both election periods and (iii) jurisdictions with a SPD majority in the first election
period and a CDU majority in the second election period.11
We use the same set of matching variables for both subsamples. Specifically, we
include (ln) real expenditures, (ln) financial expenditures and the local business tax
rate as well as the voluntary spending shares for general public goods, people public
goods, culture and infrastructure public goods in the year prior to the second-period
election. In robustness checks, we will, furthermore, change the set of matching variables
to include variables for changes in spending patterns (overall spending and spending
composition) prior to the second-period election. Further, we match on state dummies.
We target for all variables (except for the state dummies) mean and variance of the
10One advantage of entropy balancing compared to coarsened exact matching, another covariate
balancing approach, is that entropy balancing does not disregard treatment and control observations
that cannot be matched exactly.11If we observe three election periods for a particular jurisdiction, the jurisdiction could be in
both subsamples depending on the majority. Note, moreover, that similar results emerge when the
jurisdictions with stable right-wing and left-wing majorities are discarded from the control groups.
12
distribution. The resulting balancing statistics for panel A and panel B are shown in
Table (A1) and Table (A2) in the Appendix.12 The descriptive statistics suggests that
the entropy balancing is highly effective as all variables in the re-weighted sample are
very similarly distributed for treatment and control group. Further, the balancing also
decreases differences in the size of jurisdictions.
4 Results
In this section we report the results of our analysis. We start with a simple fixed ef-
fects model that includes all jurisdiction-years where the local council was dominated by
the SPD or CDU (Panel A and Panel B). The results are presented in row (1) of Table
(4). Specifically, we test for partisan effects on the log of overall real expenditures, fi-
nancial expenditures, voluntary expenditures as well as localities’ spending composition
as captured by voluntary spending shares for general expenditures, ’people-oriented’
expenditures, cultural expenditures and infrastructure spending. The results suggest
that jurisdictions with a SPD majority levy, on average, a higher tax rate on business
profits and spend slightly more on ’people-oriented’ public goods than communities
with a CDU council majority. Both effects are, however, only marginally statistically
significant. Estimates for partisan effects in the other spending dimensions, moreover,
do not point to statistically significant differences in policy choices between CDU and
SPD.
To address potential selection into treatment, we split the sample into the two sub-
samples described above, Panel A and Panel B. Row (2) and (3) show the difference-in-
differences results that compare spending trends for jurisdictions with unchanged CDU
or CDU coalition majority and jurisdictions with a change from CDU to SPD majority.
Row (4) and (5) show the difference-in-differences results that compare spending trends
for jurisdictions with unchanged SPD or SPD coalition majority and jurisdictions with
a change from SPD to CDU majority. In row (2) and (4) we use the unweighted sam-
ples and in row (3) and (5) the entropy balanced samples. When comparing spending
patterns in localities that moved from CDU majorities in the local councils to SPD
majorities with those that kept CDU/right-wing majorities throughout the two consec-
utive election periods, we find that SPD majorities significantly increase the spending
12Tables A3 and A4 show the balancing statistics when the changes in spending patterns before the
second-period election are used as matching variables.
13
share assigned to ’people-oriented’ public goods and services. This is consistent with
the baseline fixed effects result, but is more precisely estimated and the effects turn out
to be quantitatively larger. In the unweighted sample, the point estimate suggests that
the spending share for ’people-oriented’ expenditures increases by around 4 percentage
points when the SPD takes over the council majority, which is quantitatively substan-
tial given that the mean spending share is around 18% for this spending category. In
the entropy-balanced sample, the effect further increases in size to 4.6%. Moreover, the
estimates from the entropy-balanced sample also suggest that a move from a CDU to a
SPD majority is associated with lower spending on infrastructure public goods. Again,
the effect is quantitatively substantial, suggesting that infrastructure spending drops
by 3.6 percentage points. The findings, however, do not point to significant partisan
effects on (ln) overall real government spending, (ln) financial spending, (ln) overall
voluntary spending and jurisdictions’ tax rate choice. Table (A5) in the Appendix fur-
thermore shows that these findings are not driven by the definition of our voluntary
spending measures. Modelling the spending composition of our sample communities by
overall spending shares for each category (instead of voluntary spending shares) yields
similar results.
Note, again, that augmenting the estimation model by time-varying socio-economic
community controls coupled with the entropy matching implies that the estimation
approach allows for differences in underlying spending trends related to changes in
observed locality characteristics (e.g. changes in the unemployment rate, population
or age structure of the locality) as well as differences in spending trends that root in
the composition and size of government spending or tax policies prior to the second-
period election. Table A6 in the Appendix, moreover, shows that these results are
robust to changing the set of matching variables in the entropy-balancing approach
to include variables that capture changes in the public spending level and spending
structures prior to the second-period election (namely, changes from two years prior to
the second-period election to the election year). In this sample, matched communities
hence have similar pre-treatment trends in overall spending and spending composition.
Interestingly, a different result pattern emerges when we study changes from SPD
to CDU majorities. Here, the coefficient estimates for the CDU dummy turn out in-
significant across all defined spending variables if all years of the election period are
considered, cf. row (4) and (5) of Table 4. There are two potential explanations for
this result pattern. Firstly, as the baseline estimates do not identify partisan effects
for overall public good spending but only for the composition of spending, the results
14
Tab
le4:
Est
imat
ion
Res
ult
s-
Ove
rall
Sp
endin
gan
dShar
esfo
rV
olunta
rySp
endin
g
Dep
.V
ar.
(ln
)(l
n)
(ln
)L
oca
lS
har
e
Rea
lF
inan
cial
Vol
unta
ryB
usi
nes
sG
ener
al
Peo
ple
Cu
ltu
reIn
fras
tru
ctu
re
Exp
.E
xp
.E
xp
.T
axR
ate
Exp
.E
xp
.E
xp
.E
xp
.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
SP
Dvs
Un
ion
Majo
rity
inth
eL
oca
lC
ounci
l:N
=15
,614
SP
D0.
016
-0.0
090.
037
2.222
*-0
.692
1.1
32*
-0.0
41
-0.4
11
(0.0
21)
(0.0
34)
(0.0
27)
(1.2
79)
(0.5
54)
(0.6
80)
(0.1
50)
(0.8
40)
Unw
eighte
dP
an
elA
:C
DU
Coa
liti
on
-CD
UC
oali
tion
,C
DU
-CD
Uvs
CD
U-S
PD
,N
=10
,471
SP
D0.
030
0.02
60.
076
4.589
-0.2
224.5
38*
**0.
137
-2.5
78
(0.0
36)
(0.0
65)
(0.0
51)
(3.6
66)
(1.6
55)
(1.5
81)
(0.2
39)
(1.5
96)
Entr
opy-B
ala
nce
dP
an
elA
:C
DU
Coa
liti
on
-CD
UC
oali
tion
,C
DU
-CD
Uvs
CD
U-S
PD
,N
=10,
471
SP
D0.
041
0.04
50.
065
2.127
-0.0
535.0
71*
**0.
167
-3.8
11*
*
(0.0
33)
(0.0
60)
(0.0
52)
(3.1
79)
(1.1
75)
(1.3
71)
(0.1
66)
(1.5
62)
Unw
eighte
dP
an
elB
:S
PD
Coa
liti
on-S
PD
-Coal
itio
n,
SP
D-S
PD
vs.
SP
D-C
DU
,N
=4,
144
CD
U0.
027
0.04
4-0
.010
-2.0
31
-0.1
11
0.1
81
0.116
-0.2
73
(0.0
33)
(0.0
51)
(0.0
39)
(1.6
60)
(0.7
30)
(1.0
59)
(0.2
28)
(1.3
11)
Entr
opy-B
ala
nce
dP
an
elB
:SP
DC
oal
itio
n-S
PD
-Coa
liti
on,
SP
D-S
PD
vs.
SP
D-C
DU
,N
=4,
144
CD
U0.
033
0.03
40.
018
-2.2
86-0
.422
1.0
53
-0.0
60-1
.054
(0.0
34)
(0.0
52)
(0.0
45)
(1.7
20)
(0.7
73)
(1.2
50)
(0.1
58)
(1.0
72)
Con
trol
Var
,M
un
icip
alit
yan
dS
tate
-Yea
rF
EX
XX
XX
XX
X
Notes:
Tab
lesh
ow
sth
ep
oin
tes
tim
ate
sfo
rS
PD
(row
(1)
to(3
))or
CD
Um
ajo
rity
(row
(4)
an
d(5
))in
the
loca
lco
un
cil
usi
ng
diff
eren
td
epen
den
tvari
ab
les.
All
regre
ssio
nin
clu
de
contr
ol
vari
ab
les,
mu
nic
ipality
fixed
effec
tas
wel
las
state
-yea
rfi
xed
effec
ts.
Th
ed
epen
den
tvari
ab
les
are
:(l
n)
real
spen
din
g(c
ol.
(1))
,(l
n)
fin
an
cial
spen
din
g(c
ol.
(2))
,(l
n)
volu
nta
rysp
end
ing
(col.
(3))
,ju
risd
icti
on
s’b
usi
nes
sta
xra
tech
oic
e(c
ol.
(4))
an
dth
evolu
nta
rysp
end
ing
share
sfo
rgen
eral
pu
blic
goods
(col.
(5))
,p
eop
lep
ub
lic
good
s(c
ol.
(6))
,cu
ltu
re(c
ol.
(7))
an
d
infr
ast
ruct
ure
pu
blic
good
s(c
ol.
(8))
.R
ow
(1)
show
sth
eeff
ect
when
ass
um
ing
sym
met
ryfo
rth
ech
an
ge
from
SP
Dto
CD
Uan
db
ack
.In
row
(2)
an
d(3
)w
efo
cus
on
juri
sdic
tion
sw
ith
un
chan
ged
CD
Uor
CD
U-c
oaliti
on
ma
jori
tyin
the
loca
lco
un
cil
or
ach
an
ge
from
CD
Uto
SP
Dm
ajo
rity
.In
row
(4)
an
d(5
)w
efo
cus
on
juri
sdic
tion
sw
ith
eith
eru
nch
an
ged
SP
Dor
SP
D-c
oaliti
on
ma
jori
tyin
the
loca
lco
un
cil
or
ach
an
ge
from
SP
Dto
CD
Um
ajo
rity
.In
row
(2)
an
d(4
)an
unw
eighte
dsa
mp
lean
din
row
(3)
an
d(5
)an
entr
opy-b
ala
nce
dsa
mp
leis
use
d.
Sta
nd
ard
erro
rsin
pare
nth
esis
are
rob
ust
an
dcl
ust
ered
at
the
mu
nic
ipali
tyle
vel
.*,
**,
an
d***
den
ote
sign
ifica
nce
at
the
10,
5,
an
d1%
level
.Source:
Au
thors
’ca
lcu
lati
on
sb
ase
d
on
Sta
tist
ikL
okal
an
dJah
resr
ech
nu
ngss
tati
stik
1994
to2006.
15
might relate to the aggregation of individual spending items to broad spending cate-
gories. We hence re-estimated the baseline models using spending shares for individual
spending items as the dependent variable. Analogously to the baseline analysis, we
restrict our view to spending shares related to our voluntary expenditures, as defined
above, since these are the ones that can plausibly (most strongly) be influenced by par-
tisan politics. Moreover, we focus on individual spending items in the sub-categories of
’people-oriented’ goods expenditures and infrastructure spending as the baseline find-
ings supported partisan effects in these dimensions. Columns (1) to (3) of 5 reestimate
the baseline specifications with the three individual spending items in the ’people-
oriented’ expenditure category: spending for schools, spending for recreation goods
and services and spending for social goods and services (see Footnote 9 and Table 2).
Analogously, Columns (4) and (6) reestimate the baseline model for three particular
spending items related to infrastructure spending, namely spending for streets, public
facilities and economic promotion. The findings largely reject that the difference in the
partisan effects observed in Panels A and B of the baseline analysis masks (potentially
consistent) partisan effects of different sign and size in individual sub-categories.
Another explanation for the result pattern is that partisan effects are partly masked
by changes in local policies that relate to the fact that local council majorities change in
our empirical setting. If leadership changes impact on spending and tax policy choices,
this may confound our empirical estimates. To test for this possibility, we interact
the majority indicator with a continuous variable that measure the year within the
election period (e.g. first year = 0, second year = 1, etc). The approach hence allows
for differences in partisan effects across election years in our setting. If the fact that new
partisan majorities dominate the local council affects spending policies (irrespective of
whether the change is from CDU to SPD or vice versa), this may impact our estimates
for spending in post-election years.
The results are presented in Table (6).13 While both, changes from CDU to SPD
majorities and changes from SPD to CDU majorities are not associated with differences
in the level of real, financial or voluntary spending, the results suggest that both SPD
and CDU, when they newly gain the majority in the local council, adjust the compo-
sition of public spending and allocate more spending to ’people-oriented’ public goods
and services and less towards infrastructure spending relative to councils where the
dominant party remains unchanged in the election. This effect declines over the elec-
13Observations are unweighted in all specifications of 6. Similar results, however, are obtained in
entropy-balanced samples.
16
Table 5: Estimation Results - Shares for Voluntary Spending in Subcategories
Dep. Var. Expenditure Shares
’People-Oriented’ Exp. Infrastructure Exp.
Schools Recreation Social Streets Public Economic
Facilities Promotion
(1) (2) (3) (4) (5) (6)
SPD vs Union Majority in the Local Council: N=15,833
SPD -0.143 0.637* 0.638* -0.244 -0.108 -0.058
(0.366) (0.347) (0.364) (0.687) (0.347) (0.284)
Panel A: CDU Coalition-CDU Coalition, CDU-CDU vs CDU-SPD: N = 10,471
Unweighted Panel A
SPD 2.028** 0.438 2.073** -0.921 -0.936* -0.720***
(0.862) (0.550) (0.916) (1.647) (0.512) (0.238)
Entropy-Balanced Panel A
SPD 2.128** 0.858* 1.664** -1.282 -1.474** -0.873**
(0.864) (0.501) (0.824) (1.484) (0.606) (0.367)
Panel B: SPD Coalition-SPD-Coalition, SPD-SPD vs. SPD-CDU: N = 4,144
Unweighted Panel B
CDU 0.556 -0.097 -0.277 -0.822 0.340 0.209
(0.535) (0.543) (0.595) (0.981) (0.644) (0.388)
Entropy-Balanced Panel B
CDU 1.336* -0.347 0.064 -1.184 0.228 -0.099
(0.756) (0.676) (0.601) (0.783) (0.599) (0.356)
Control Variables X X X X X X
Municipality FE X X X X X X
State-Year FE X X X X X X
Notes: Table shows the point estimates for SPD (row (1) to (3)) or CDU majority (row (4) and (5)) in the local
council using different dependent variables. All regression include control variables, municipality fixed effect as well
as state-year fixed effects. The dependent variables are: (ln) real spending (col. (1)), (ln) financial spending (col. (2)),
(ln) voluntary spending (col. (3)), jurisdictions’ local business tax rate choice (col. (4)) and the voluntary spending
shares for general public goods (col. (5)), people public goods (col. (6)), culture (col. (7)) and infrastructure public
goods (col. (8)). Row (1) shows the effect when assuming symmetry for the change from SPD to CDU and back.
In row (2) and (3) we focus on jurisdictions with unchanged CDU or CDU-coalition majority in the local council
or a change from CDU to SPD majority. In row (4) and (5) we focus on jurisdictions with either unchanged SPD
or SPD-coalition majority in the local council or a change from SPD to CDU majority. In row (2) and (4) an
unweighted sample and in row (3) and (5) an entropy-balanced sample is used. Standard errors in parenthesis are
robust and clustered at the municipality level. *, **, and *** denote significance at the 10, 5, and 1% level. Source:
Authors’ calculations based on Statistik Lokal and Jahresrechnungsstatistik 1994 to 2006.17
toral period. In later years, the results are consistent with symmetric partisan effects in
both samples (Panel A and Panel B): If the SPD gains the council majority, spending
on ’people-oriented’ goods is increased at the expense of spending on infrastructure; if
the CDU gains the council majority, spending on infrastructure is increased relative to
spending on ’people-oriented’ goods. See also the specifications in the last row of Table
6 which compares spending patterns between SPD and CDU dominated local councils
in the last two years of the electoral cycle.14 Quantitatively, the latter estimates suggest
that the SPD dominated councils spend around 2%-points more on ’people-oriented’
public goods and 2% less on infrastructure public goods (significant at the 5% and
10% level). In relative terms this means between 10 and 15% more (less) spending on
people public goods (infrastructure public goods). The driving subcategories (not re-
ported) are recreation (1.7, significant at the 5% level) and social spending (1%-points,
significant at the 5% level) as well as spending for streets (-1.3%-points, p-value: 0.16).
Our results hence suggest that changes in council majorities (irrespective of the
direction of the partisan change) may impact on spending composition. Specifically,
both, CDU and SPD are suggested to spend more on people goods and less on infras-
tructure goods when they gain power and take over the majority of the local council
compared to communities where the dominant party in the council remains unchanged.
One potential theoretical explanation for this effect is that opposition parties, in elec-
toral campaigns, have to differentiate from incumbent policy positions to sharpen their
profile and gain medial visibility. Opposition parties might find it attractive to promise
and, if elected into power, eventually implement spending policies that put more em-
phasis on consumptive spending (like spending for the ’people-oriented’ goods and
services) than on investment spending (like infrastructure spending) as the latter goes
down better with short-sighted voters. On top of that, the observed pattern might
reflect responses to incumbents’ pre-election spending choices, which tend to be char-
acterized by increased investment spending in pre-election years (and constant overall
spending, see e.g. Kneebone and McKenzie (2001), Gonzalez (2002), Brender (2003),
Khemani (2004) and Drazen and Eslava (2010)). Our results are consistent with the
notion that parties that gain legislative power reverse these spending trends (more
strongly than incumbent ones) after the election.
14Note that Table (A7) in the Appendix, moreover, presents results from specifications, where timing
effects across the electoral cycle are modelled in a more flexible non-linear way.
18
Tab
le6:
Est
imat
ion
Res
ult
s-
Tim
ing
ofSp
endin
g
Dep
.V
ar.
(ln
)(l
n)
(ln
)L
oca
lS
har
e
Rea
lF
inan
cial
Vol
unta
ryB
usi
nes
sG
ener
al
Peo
ple
Cult
ure
Infr
astr
uct
ure
Exp
.E
xp
.E
xp
.T
axR
ate
Exp
.E
xp
.E
xp
.E
xp
.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Pan
elA
:C
DU
Coa
liti
on
-CD
UC
oal
itio
n,
CD
U-C
DU
vs
CD
U-S
PD
,N
=10
,098
SP
D0.
020
0.03
00.
022
1.53
6-0
.259
7.0
52*
-0.1
20-6
.273
***
(0.0
47)
(0.0
64)
(0.0
64)
(2.3
15)
(2.1
95)
(3.6
60)
(0.2
01)
(1.8
19)
SP
D*
#Y
ear
0.0
04-0
.001
0.01
91.
074
0.0
13
-0.8
850.0
90
1.3
00*
(0.0
13)
(0.0
20)
(0.0
15)
(1.3
40)
(0.4
51)
(0.9
88)
(0.1
23)
(0.6
96)
Pan
elB
:S
PD
Coa
liti
on
-SP
D-C
oali
tion
,S
PD
-SP
Dvs.
SP
D-C
DU
,N
=4,
144
CD
U0.
013
0.05
8-0
.037
-1.6
720.5
52
2.272
*-0
.620
-2.7
57*
(0.0
37)
(0.0
58)
(0.0
49)
(1.6
01)
(1.0
65)
(1.2
73)
(0.3
85)
(1.5
56)
CD
U*
#Y
ear
0.0
05-0
.005
0.01
0-0
.129
-0.2
39
-0.7
55**
0.2
66*
*0.8
97*
(0.0
09)
(0.0
18)
(0.0
13)
(0.6
87)
(0.2
48)
(0.3
52)
(0.1
16)
(0.5
00)
SP
Dvs
Un
ion
Majo
rity
inth
eL
oca
lC
oun
cil,
only
last
two
year
sin
anel
ecti
onp
erio
d:
N=
6,7
46
SP
D-0
.022
-0.0
36-0
.009
1.83
0-0
.442
2.082
**0.
070
-1.9
63*
(0.0
23)
(0.0
37)
(0.0
31)
(1.4
41)
(0.7
14)
(0.9
48)
(0.1
91)
(1.0
16)
Contr
ol
Var
iab
les,
Mu
nic
ipal
ity
FE
XX
XX
XX
XX
an
dS
tate
-Yea
rF
EX
XX
XX
XX
X
Notes:
Tab
lesh
ow
sth
ep
oin
tes
tim
ate
sfo
rS
PD
ma
jori
tyin
the
loca
lco
un
cil
usi
ng
diff
eren
td
epen
den
tvari
able
s.A
llre
gre
ssio
nin
clu
de
contr
ol
vari
ab
les,
mun
icip
ality
fixed
effec
tas
wel
las
state
-yea
rfi
xed
effec
ts.
Th
ed
epen
den
tvari
ab
les
are
:(l
n)
real
spen
din
g(c
ol.
(1))
,(l
n)
fin
an
cial
spen
din
g(c
ol.
(2))
,(l
n)
volu
nta
rysp
end
ing
(col.
(3))
an
dju
risd
icti
on
s’lo
cal
bu
sin
ess
tax
rate
choic
e(c
ol.
(4))
.F
urt
her
,in
col.
(5)
to(8
)it
isth
evolu
nta
rysp
end
ing
share
of
gen
eral
pu
blic
good
s,p
eop
lep
ub
lic
good
s,cu
ltu
rean
din
frast
ruct
ure
pu
blic
good
s.In
row
(1)
we
focu
son
juri
sdic
tion
sw
ith
un
chan
ged
CD
Uor
CD
U-c
oali
tion
ma
jori
ty
inth
elo
cal
cou
nci
lor
ach
an
ge
from
CD
Uto
SP
Dm
ajo
rity
.In
row
(2)
we
focu
son
juri
sdic
tion
sw
ith
eith
eru
nch
an
ged
SP
Dor
SP
D-c
oaliti
on
ma
jori
tyin
the
loca
l
cou
nci
lor
ach
an
ge
from
SP
Dto
CD
Um
ajo
rity
.In
row
(3)
we
incl
ud
eall
juri
sdic
tion
sw
ith
eith
erS
PD
or
CD
Um
ajo
rity
bu
ton
lyth
ela
sttw
oel
ecti
on
per
iod
yea
rs.
Sta
nd
ard
erro
rsin
pare
nth
esis
are
rob
ust
an
dcl
ust
ered
at
the
mu
nic
ipality
level
.*,
**,
an
d***
den
ote
sign
ifica
nce
at
the
10,
5,
an
d1%
level
.Source:
Au
thors
’ca
lcu
lati
on
sb
ase
don
Sta
tist
ikL
okal
an
dJah
resr
ech
nu
ngss
tati
stik
1994
to2006.
19
5 Conclusion
The aim of this paper was to assess the role of partisanship of West German local
council majorities on overall local public spending and the composition of local pub-
lic spending. We combine a simple fixed effect regression approach with an entropy
balanced matching strategy to empirically identify the effect of interest. Two find-
ings emerge: Firstly, our results point to sizable partisan effects on the composition of
public spending (while aggregate spending levels are suggested to remain unaffected).
Secondly, our findings suggest that the composition of public spending changes sys-
tematically when parties gain the majority of seats in legislative bodies. This, from
our perspective, is an important insight as it affects empirical identification strategies
of partisan effects based on election-induced changes in partisanship (irrespective of
the methodological identification approach - it e.g. emerges in difference-in-differences
strategies as well as in regression discontinuity designs).
20
References
Alesina, A., 1988. Credibility and policy convergence in a two-party system with ra-
tional voters. The American Economic Review 78, pp. 796–805.
Besley, T., Case, A., 2003. Political institutions and policy choices: Evidence from the
united states. Journal of Economic Literature 41, pp. 7–73.
Besley, T., Coate, S., 1997. An economic model of representative democracy. The Quar-
terly Journal of Economics 112, 85–114.
Brender, A., 2003. The effect of fiscal performance on local government election results
in israel: 1989?1998. Journal of Public Economics, 2187–2205.
Brender, A., Drazen, A., 2013. Elections, leaders, and the composition of government
spending. Journal of Public Economics, 18–31.
Drazen, A., Eslava, M., 2010. Electoral manipulation via voter-friendly spending: the-
ory and evidence. Journal of Development Economics, 39–52.
Ferreira, F., Gyourko, J., 2009. Do political parties matter? evidence from u.s. cities.
The Quarterly Journal of Economics 124, pp. 399–422.
Freier, R., Odendahl, C., 2015. Do parties matter? estimating the effect of political
power in multi-party systems. European Economic Review 80, 310 – 328.
Glaeser, E. L., Ponzetto, G. A. M., Shapiro, J. M., 2005. Strategic extremism: Why
republicans and democrats divide on religious values. The Quarterly Journal of Eco-
nomics 120, 1283–1330.
Gonzalez, M., 2002. Do changes in democracy affect the political budget cycle? evidence
from mexico. Review of Development Economics, 204–224.
Hainmueller, J., 2012. Entropy balancing for causal effects: A multivariate reweighting
method to produce balanced samples in observational studies. Political Analysis 20,
25–46.
Imbeau, L. M., Petry, F., Lamari, M., 2001. Left-right party ideology and government
policies: A meta-analysis. European Journal of Political Research 40, 1–29.
Imbens, G. W., 2004. Nonparametric estimation of average treatment effects under
exogeneity: A review. The Review of Economics and Statistics 86, 4–29.
21
Khemani, S., 2004. Political cycles in a developing economy: effect of elections in the
indian states. Journal of Development Economics, 125–154.
Kneebone, R., McKenzie, K., 2001. Electoral and partisan cycles in fiscal policy: an
examination of canadian provinces. International Tax and Public Finance, 753–774.
Lee, D. S., Moretti, E., Butler, M. J., 2004. Do voters affect or elect policies? evidence
from the u. s. house. The Quarterly Journal of Economics 119, 807–859.
Pappi, F. U., Eckstein, G., 1998. Voters’ party preferences in multiparty systems and
their coalitional and spatial implications: Germany after unification. Public Choice
97, 229–255.
Potrafke, N., 2011a. Does government ideology influence budget composition? empirical
evidence from oecd countries. Economics of Governance 12, 101–134.
Potrafke, N., 2011b. Public expenditures on education and cultural affairs in the west
german states: Does government ideology influence the budget composition? German
Economic Review 12, 124–145.
Sole-Olle, A., Viladecans-Marsal, E., 2013. Do political parties matter for local land
use policies? Journal of Urban Economics 78, 42 – 56.
22
Appendix: Additional Descriptive Statistics and Es-
timation Results
Table A1: Descriptive Statistics for Panel A: Unweighted and Entropy-Balanced
Control Treatment
CDU-Coalition-CDU-Coalition, CDU-CDU CDU-SPD
Unweighted Entropy-Balanced
Mean Variance Mean Variance Mean Variance
(ln) Population in 1000 1.62 1.08 0.92 0.42 0.84 0.41
Share population under 20 0.23 0.00 0.23 0.00 0.23 0.00
Share population over 65 0.17 0.00 0.17 0.00 0.17 0.00
(ln) Employees 7.38 1.09 6.68 0.46 6.65 0.45
Unemployment rate (county) 8.50 3.25 8.68 2.86 9.29 1.93
Debt per capita (county) 6.75 0.11 6.85 0.07 6.86 0.03
Local business tax multiplier 339.91 1362.63 326.82 345.69 326.80 345.03
(ln) Real expenditures 15.91 1.24 15.14 0.47 15.12 0.45
(ln) Financial expenditures 15.03 1.25 14.38 0.50 14.38 0.50
Voluntary expenditure shares for
Share general public good expenditures 12.77 21.03 13.15 11.79 13.15 11.78
Share people public good expenditures 18.09 86.28 23.87 138.96 23.87 138.95
Share culture expenditures 1.30 3.78 0.71 0.42 0.70 0.41
Share firm public good expenditures 17.45 72.29 18.42 131.16 18.42 131.17
Observations 1,121 1,121 15
Notes: Table reports descriptive statistics for jurisdictions with CDU dominated councils that
changed to a SPD dominated council (treatment group) and jurisdictions with CDU or CDU-
Coalition dominated councils with no change in the majority in the local council, before entropy-
balancing (unweighted) and after. The variables refer to the second last period before the next
election. Every jurisdiction is counted only once. The matching variables are local business tax mul-
tiplier, (ln) real and financial spending as well as the voluntary expenditure shares. Source: Authors’
calculations based on Statistik Lokal and Jahresrechnungsstatistik 1994 to 2006.
23
Table A2: Descriptive Statistics for Panel B: Unweighted and Entropy-Balanced
Control Treatment
SPD-Coalition-SPD-Coalition, SPD-SPD SPD-CDU
Unweighted Entropy-Balanced
Mean Variance Mean Variance Mean Variance
(ln) Population in 1000 1.53 1.08 1.79 0.91 1.71 0.83
Share population under 20 0.22 0.00 0.22 0.00 0.22 0.00
Share population over 65 0.17 0.00 0.17 0.00 0.18 0.00
(ln) Employees 7.33 1.11 7.58 0.93 7.47 0.80
Unemployment rate (county) 9.53 4.43 9.50 2.85 9.59 3.23
Debt per capita (county) 6.89 0.11 6.71 0.11 6.77 0.07
Local business tax multiplier 332.90 1063.89 342.79 1028.86 342.77 1029.48
(ln) Real expenditures 15.80 1.20 16.01 0.94 16.01 0.94
(ln) Financial expenditures 14.91 1.19 15.05 1.00 15.05 1.00
Voluntary expenditure shares for
General public goods 13.35 14.78 13.40 9.46 13.40 9.46
People public goods 18.67 75.34 17.02 103.34 17.02 103.43
Culture 1.39 4.96 1.12 1.69 1.12 1.69
Infrastructure public goods 16.99 50.21 13.29 34.16 13.29 34.17
Observations 399 399 44
Notes: Table reports descriptive statistics for jurisdictions with SPD dominated councils that changed
to a CDU dominated council (treatment group) and jurisdictions with SPD or SPD-Coalition dom-
inated councils with no change in the majority in the local council, before entropy-balancing (un-
weighted) and after. The variables refer to the last year before the next election. Every jurisdiction
is counted only once. The matching variables are local business tax multiplier, (ln) real and finan-
cial spending as well as the voluntary expenditure shares. Source: Authors’ calculations based on
Statistik Lokal and Jahresrechnungsstatistik 1994 to 2006.
24
Table A3: Descriptive Statistics for Panel A: Unweighted and Entropy-Balanced:
Matching based on Changes in Matching Variables
Control Treatment
CDU-Coalition-CDU-Coalition, CDU-CDU CDU-SPD
Unweighted Entropy-Balanced
Mean Variance Mean Variance Mean Variance
(ln) Population in 1000 1.62 1.07 1.21 0.79 0.84 0.41
Share population under 20 0.23 0.00 0.23 0.00 0.23 0.00
Share population over 65 0.17 0.00 0.17 0.00 0.17 0.00
(ln) Employees 7.39 1.09 6.96 0.85 6.65 0.45
Unemployment rate (county) 8.50 3.26 8.85 4.86 9.29 1.93
Debt per capita (county) 6.75 0.11 6.81 0.08 6.86 0.03
∆ Local business tax multiplier 6.10 119.64 3.00 27.93 3.00 27.86
∆ (ln) Real expenditures 0.02 0.04 0.10 0.04 0.10 0.04
∆ (ln) Financial expenditures 0.02 0.07 0.12 0.04 0.12 0.04
Voluntary expenditure shares for
∆ General public goods 0.19 17.10 -2.43 25.47 -2.43 25.47
∆ People public goods 0.27 60.65 0.59 55.17 0.59 55.20
∆ Culture -0.03 3.42 0.00 0.65 0.00 0.65
∆ Infrastructure public goods 0.55 80.69 3.25 89.07 3.25 89.14
Observations 1,118 1,118 15
Notes: Table reports descriptive statistics for jurisdictions with CDU dominated councils that
changed to a SPD dominated council (treatment group) and jurisdictions with CDU or CDU-
Coalition dominated councils with no change in the majority in the local council, before entropy-
balancing (unweighted) and after. The variables refer to the last period before the next election. ∆
measure the two-year change. Every jurisdiction is counted only once. The matching variables are
the two year changes of (ln) real expenditures, (ln) financial expenditures and the local business
tax multiplier as well as the spending share for general, people, culture and infrastructure public
goods in the last year before the election. Source: Authors’ calculations based on Statistik Lokal and
Jahresrechnungsstatistik 1994 to 2006.
25
Table A4: Descriptive Statistics for Panel B: Unweighted and Entropy-Balanced Match-
ing based on Changes in Matching Variables
Control Treatment
SPD-Coalition-SPD-Coalition, SPD-SPD SPD-CDU
Unweighted Entropy-Balanced
Mean Variance Mean Variance Mean Variance
(ln) Population in 1000 1.54 1.08 1.71 1.45 1.71 0.83
Share population under 20 0.22 0.00 0.21 0.00 0.22 0.00
Share population over 65 0.17 0.00 0.17 0.00 0.18 0.00
(ln) Employees 7.33 1.11 7.50 1.48 7.47 0.80
Unemployment rate (county) 9.53 4.44 9.76 4.93 9.59 3.23
Debt per capita (county) 6.89 0.11 6.71 0.09 6.77 0.07
∆ Local business tax multiplier 2.26 43.99 4.21 58.16 4.20 58.07
∆ (ln) Real expenditures 0.02 0.03 0.01 0.03 0.01 0.03
∆ (ln) Financial expenditures 0.04 0.07 0.04 0.08 0.04 0.08
Voluntary expenditure shares for
∆ General public goods 0.32 12.71 0.13 8.33 0.13 8.34
∆ People public goods 0.38 36.21 1.79 58.17 1.79 58.20
∆ Culture 0.01 2.77 -0.17 1.17 -0.17 1.17
∆ Infrastructure public goods 0.27 59.66 -1.24 26.61 -1.24 26.61
Observations 399 399 44
Notes: Table reports descriptive statistics for jurisdictions with SPD dominated councils that changed
to a CDU dominated council (treatment group) and jurisdictions with SPD or SPD-Coalition dom-
inated councils with no change in the majority in the local council, before entropy-balancing (un-
weighted) and after. The matching variables are (ln) real expenditures, (ln) financial expenditures
and the local business tax multiplier as well as the spending share for general, people, culture and
infrastructure public goods in the last year before the election. The variables refer to the last year
before the next election. ∆ measure the two-year change. Every jurisdiction is counted only once.
The matching variables are the two year changes of (ln) real expenditures, (ln) financial expenditures
and the local business tax multiplier as well as the spending share for general, people, culture and
infrastructure public goods in the last year before the election. Source: Authors’ calculations based
on Statistik Lokal and Jahresrechnungsstatistik 1994 to 2006.
26
Table A5: Estimation Results - Overall Spending Shares (Mandatory and Voluntary
Spending)
Dep. Var. Share
General People Culture Infrastructure
Expenditure Expenditure Expenditure Expenditure
(1) (2) (3) (4)
SPD vs CDU Majority in the Local Council: N=15,834
SPD -1.215* 1.482* -0.142 -0.192
(0.625) (0.788) (0.174) (0.945)
CDU Coalition-CDU Coalition, CDU-CDU vs CDU-SPD, N = 10,098
Unweighted Panel A
SPD -0.350 3.795** 0.036 -3.792*
(1.467) (1.877) (0.216) (2.132)
Entropy-Balanced Panel A
SPD -0.174 4.093** 0.100 -4.359**
(0.917) (1.694) (0.156) (1.916)
SPD Coalition-SPD-Coalition, SPD-SPD vs. SPD-CDU, N = 4,144
Unweighted Panel B
CDU 1.001 -1.073 0.287 -0.248
(0.980) (1.165) (0.278) (1.352)
Entropy-Balanced Panel B
CDU 1.072 -0.540 0.004 -0.643
(1.055) (1.334) (0.175) (1.213)
Control Variables X X X X
Municipality FE X X X X
State-Year FE X X X X
Notes: Table shows the point estimates for SPD (row (1) to (3)) or CDU majorities (row (4) and (5)) in the local
council using different dependent variables. All regression include control variables, municipality fixed effect as well
as state-year fixed effects. The dependent variables are voluntary expenditures share for general public goods (col.
(1)), people public goods (col. (2)), culture (col. (3)) and infrastructure public goods (col. (4)). Row (1) shows
the effect when assuming symmetry for the change from SPD to CDU and back. In row (2) and (3) we focus on
jurisdictions with unchanged CDU or CDU-coalition majority in the local council or a change from CDU to SPD
majority. In row (4) and (5) we focus on jurisdictions with either unchanged SPD or SPD-coalition majority in the
local council or a change from SPD to CDU majority.In row (2) and (4) an unweighted sample and in row (3) and
(5) an entropy-balanced sample is used. Standard errors in parenthesis are robust and clustered at the municipality
level. *, **, and *** denote significance at the 10, 5, and 1% level. Source: Authors’ calculations based on Statistik
Lokal and Jahresrechnungsstatistik 1994 to 2006. 27
Tab
leA
6:E
stim
atio
nR
esult
s-
Mat
chin
gon
Chan
ges
ofM
atch
ing
Var
iable
s
Dep
.V
ar.
(ln
)(l
n)
(ln
)L
oca
lS
har
e
Rea
lF
inan
cial
Vol
unta
ryB
usi
nes
sG
ener
al
Peo
ple
Cu
ltu
reIn
fras
tru
ctu
re
Exp
.E
xp
.E
xp
.T
axR
ate
Exp
.E
xp
.E
xp
.E
xp
.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Entr
opy-B
ala
nce
dP
an
elA
:C
DU
Coa
liti
on
-CD
UC
oali
tion
,C
DU
-CD
Uvs
CD
U-S
PD
,N
=10,
071
SP
D0.
011
-0.0
020.
063
4.650
1.6
56
4.216
***
0.1
11
-3.7
40**
*
(0.0
31)
(0.0
58)
(0.0
49)
(3.1
71)
(1.2
14)
(1.1
97)
(0.1
65)
(1.3
75)
Entr
opy-B
ala
nce
dP
an
elB
:SP
DC
oal
itio
n-S
PD
-Coa
liti
on,
SP
D-S
PD
vs.
SP
D-C
DU
,N
=4,
128
CD
U0.
005
0.00
8-0
.011
-2.5
80*
0.8
99
-0.1
54
0.1
20
-1.2
12
(0.0
32)
(0.0
50)
(0.0
38)
(1.5
33)
(0.8
89)
(1.0
23)
(0.1
31)
(1.0
03)
Con
trol
Var
,M
un
icip
alit
yan
dS
tate
-Yea
rF
EX
XX
XX
XX
X
Notes:
Tab
lesh
ow
sth
ep
oin
tes
tim
ate
sfo
rS
PD
(row
(1))
or
CD
Um
ajo
rity
(row
(2))
inth
elo
cal
cou
nci
lu
sin
gd
iffer
ent
dep
end
ent
vari
ab
les.
All
regre
ssio
nin
clu
de
contr
ol
vari
ab
les,
mu
nic
ipality
fixed
effec
tas
wel
las
state
-yea
rfi
xed
effec
ts.
Th
ed
epen
den
tvari
ab
les
are
:(l
n)
real
spen
din
g(c
ol.
(1))
,(l
n)
fin
an
cial
spen
din
g(c
ol.
(2))
,(l
n)
volu
nta
rysp
end
ing
(col.
(3))
,
juri
sdic
tion
s’ta
xra
tech
oic
e(c
ol.
(4))
an
dth
evolu
nta
rysp
end
ing
share
sfo
rgen
eral
pu
blic
good
s(c
ol.
(5))
,p
eop
lep
ub
lic
good
s(c
ol.
(6))
,cu
ltu
re(c
ol.
(7))
an
din
frast
ruct
ure
pu
bli
c
good
s(c
ol.
(8))
.In
row
(2)
we
focu
son
juri
sdic
tion
sw
ith
un
chan
ged
CD
Uor
CD
U-c
oaliti
on
ma
jori
tyin
the
loca
lco
un
cil
or
ach
an
ge
from
CD
Uto
SP
Dm
ajo
rity
.In
row
(2)
we
focu
son
juri
sdic
tion
sw
ith
eith
eru
nch
an
ged
SP
Dor
SP
D-c
oali
tion
ma
jori
tyin
the
loca
lco
un
cil
or
ach
an
ge
from
SP
Dto
CD
Um
ajo
rity
.In
all
spec
ifica
tion
we
use
an
entr
opy-b
ala
nce
dsa
mp
le.
Th
em
ain
diff
eren
ceto
ou
rm
ain
resu
ltis
that
matc
hin
gis
don
eu
sin
gch
an
ges
inth
em
atc
hin
gvari
ab
les.
Sta
nd
ard
erro
rsin
pare
nth
esis
are
rob
ust
an
dcl
ust
ered
at
the
mu
nic
ipali
ty
level
.*,
**,
an
d***
den
ote
sign
ifica
nce
at
the
10,
5,
an
d1%
level
.Source:
Au
thors
’ca
lcu
lati
on
sb
ase
don
Sta
tist
ikL
okal
an
dJah
resr
ech
nu
ngss
tati
stik
1994
to2006.
28
Tab
leA
7:E
stim
atio
nR
esult
s-
Tim
ing
ofSp
endin
g(F
lexib
leF
unct
ional
For
ms)
Dep
.V
ar.
(ln
)(l
n)
(ln
)L
oca
lS
har
e
Rea
lF
inan
cial
Vol
unta
ryB
usi
nes
sG
ener
al
Peo
ple
Cu
ltu
reIn
frast
ruct
ure
Exp
.E
xp
.E
xp
.T
axR
ate
Exp
.E
xp
.E
xp
.E
xp
.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Pan
elA
:C
DU
Coa
liti
on
-CD
UC
oal
itio
n,
CD
U-C
DU
vs
CD
U-S
PD
,N
=10
,098
SP
D*
Ele
ctio
nan
d1s
tyea
r0.
032
0.02
50.
035
2.37
4*-0
.606
6.3
03*
*-0
.140
-5.3
92*
**
(0.0
35)
(0.0
52)
(0.0
59)
(1.4
38)
(1.4
65)
(2.8
22)
(0.1
43)
(1.0
98)
SP
D*
2n
dan
d3rd
yea
r0.0
150.
028
0.03
42.
056
0.4
02
2.967
0.1
93
-2.9
34*
(0.0
33)
(0.0
48)
(0.0
44)
(2.4
38)
(2.0
18)
(1.8
74)
(0.1
66)
(1.6
58)
SP
D*>
3rd
year
0.03
40.
014
0.10
1**
4.40
4-0
.177
1.474
0.3
55
0.126
(0.0
33)
(0.0
62)
(0.0
43)
(4.0
18)
(1.1
51)
(1.6
99)
(0.4
57)
(2.0
20)
Pan
elB
:S
PD
Coa
liti
on
-SP
D-C
oali
tion
,S
PD
-SP
Dvs.
SP
D-C
DU
,N
=4,
144
CD
U*
Ele
ctio
nan
d1st
year
-0.0
090.
031
-0.0
57-1
.018
0.1
02
1.3
62
-0.4
90*
-2.1
44
(0.0
32)
(0.0
53)
(0.0
43)
(1.1
21)
(0.8
83)
(0.9
72)
(0.2
54)
(1.3
22)
CD
U*
2nd
and
3rd
yea
r0.0
260.
057
-0.0
07-1
.676
0.6
50
0.9
51
0.004
-1.0
53
(0.0
26)
(0.0
40)
(0.0
35)
(1.1
03)
(0.6
67)
(0.9
43)
(0.3
26)
(1.0
79)
CD
U*>
3rd
year
0.0
130.
012
0.00
6-1
.122
-0.4
73-0
.892
0.379
*0.
933
(0.0
29)
(0.0
50)
(0.0
37)
(2.0
25)
(0.6
48)
(1.0
04)
(0.2
04)
(1.3
95)
Contr
ol
Var
iab
les,
Mu
nic
ipali
tyF
EX
XX
XX
XX
X
an
dS
tate
-Yea
rF
EX
XX
XX
XX
X
Notes:
Tab
lesh
ow
sth
ep
oin
tes
tim
ate
sfo
rS
PD
(row
(1)
to(3
))or
CD
Um
ajo
rity
(row
(4)
an
d(5
))in
the
loca
lco
un
cil
usi
ng
diff
eren
td
epen
den
tvari
ab
les.
All
regre
ssio
n
incl
ud
eco
ntr
ol
vari
ab
les,
mu
nic
ipality
fixed
effec
tas
wel
las
state
-yea
rfi
xed
effec
ts.
Th
ed
epen
den
tvari
ab
les
are
:(l
n)
real
spen
din
g(c
ol.
(1))
,(l
n)
fin
an
cial
spen
din
g
(col.
(2))
,(l
n)
volu
nta
rysp
end
ing
(col.
(3))
,ju
risd
icti
on
s’ta
xra
tech
oic
e(c
ol.
(4))
an
dth
evolu
nta
rysp
end
ing
share
sfo
rgen
eral
pu
blic
good
s(c
ol.
(5))
,p
eop
lep
ub
lic
good
s(c
ol.
(6))
,cu
ltu
re(c
ol.
(7))
an
din
frast
ruct
ure
pu
blic
good
s(c
ol.
(8))
.In
row
(1)
we
focu
son
juri
sdic
tion
sw
ith
un
chan
ged
CD
Uor
CD
U-c
oaliti
on
ma
jori
tyin
the
loca
lco
un
cil
or
ach
an
ge
from
CD
Uto
SP
Dm
ajo
rity
.In
row
(2)
we
focu
son
juri
sdic
tion
sw
ith
eith
eru
nch
an
ged
SP
Dor
SP
D-c
oaliti
on
ma
jori
tyin
the
loca
lco
un
cil
or
ach
an
ge
from
SP
Dto
CD
Um
ajo
rity
.S
tan
dard
erro
rsin
pare
nth
esis
are
rob
ust
an
dcl
ust
ered
at
the
mu
nic
ipality
level
.*,
**,
an
d***
den
ote
sign
ifica
nce
at
the
10,
5,
an
d1%
level
.Source:
Au
thors
’ca
lcu
lati
on
sb
ase
don
Sta
tist
ikL
okal
an
dJah
resr
ech
nu
ngss
tati
stik
1994
to2006.
29